No-reference quality assessment for image-based assessment of economically important tropical woods.

Image Quality Assessment (IQA) is essential for the accuracy of systems for automatic recognition of tree species for wood samples. In this study, a No-Reference IQA (NR-IQA), wood NR-IQA (WNR-IQA) metric was proposed to assess the quality of wood images. Support Vector Regression (SVR) was trained using Generalized Gaussian Distribution (GGD) and Asymmetric Generalized Gaussian Distribution (AGGD) features, which were measured for wood images. Meanwhile, the Mean Opinion Score (MOS) was obtained from the subjective evaluation. This was followed by a comparison between the proposed IQA metric, WNR-IQA, and three established NR-IQA metrics, namely Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE), deepIQA, Deep Bilinear Convolutional Neural Networks (DB-CNN), and five Full Reference-IQA (FR-IQA) metrics known as MSSIM, SSIM, FSIM, IWSSIM, and GMSD. The proposed WNR-IQA metric, BRISQUE, deepIQA, DB-CNN, and FR-IQAs were then compared with MOS values to evaluate the performance of the automatic IQA metrics. As a result, the WNR-IQA metric exhibited a higher performance compared to BRISQUE, deepIQA, DB-CNN, and FR-IQA metrics. Highest quality images may not be routinely available due to logistic factors, such as dust, poor illumination, and hot environment present in the timber industry. Moreover, motion blur could occur due to the relative motion between the camera and the wood slice. Therefore, the advantage of WNR-IQA could be seen from its independency from a "perfect" reference image for the image quality evaluation.

Introduction

Wood is a plant tissue consisting of a porous and fibrous structure. It is widely used as a source of energy, and for furniture making, millwork, flooring, building construction, and paper production [1]. Thousands of wood producing tree species are present, which comprise materials of distinct physical characteristics in terms of structure, density, colour, and texture [2]. These characteristics define the preferred usages and monetary values of the trees [3]. Furthermore, although the timber production at high latitudes is based on a small number of species, a wide range of tropical forests is present. For example, conifers of the genus Pine are widespread in the Northern Hemisphere. Subsequently, this phenomenon leads to the production of moderate-priced wood of high resin content, which is widely used for the making of indoor furniture.

Discovered in the native of Central America, Bocote (Cordia gerascanthus) is used to produce high-cost hardwood, which is suitable for high-quality furniture and cabinetry due to the glossy finish created by the oily surface of the wood. Meanwhile, the rosewood (Dalbergia sp.) is another high-cost wood, which is sought for instrument making and flooring due to its high strength and density. Provided that each wood species consists of various price and characteristics, misclassification of the wood could lead to financial losses. Therefore, the correct identification of the different wood species is essential.

Although the recognition of wood species is traditionally performed by humans [4], the process of it is time-consuming and incurs a high cost to the lumber industry. Therefore, various algorithms have been developed for automatic recognition of wood samples [1, 2, 5, 6]. A scope is present for the improvement in the accuracy of automatic wood recognition systems through high-quality microscopy images, which are sometimes pre-processed to enhance the recognition. However, the processes of image enhancement require more time and may impart a checkerboard artefact to the wood images [7]. Besides, the environment of timber factories is surrounded by dust, poor illumination, and heat [8], which lead to the degradation of the image quality. Therefore, a suitable Image Quality Assessment (IQA) metric is essential to evaluate the captured images before proceeding to the pipeline for recognition algorithms.

Image Quality Assessment (IQA) may be specified into two categories, namely subjective and objective evaluations. Subjective evaluation occurs when the images are evaluated by human, who provide scores based on their perception on the image quality, while objective evaluation involves mathematical algorithms to calculate the quality score for the images [9]. Although subjective evaluation is regarded as the gold standard in IQA, it is not practical in the industrial setting due to the high cost and long duration required. Therefore, an incentive is made to develop objective evaluation procedures of the comparable quality to subjective IQA evaluation [9].

The objective evaluation consists of three categories, namely Full-Reference-IQA (FR-IQA), Reduced Reference-IQA (RR-IQA), and No-Reference/Blind-IQA (NR-IQA) [10, 11]. Specifically, FR-IQA evaluates an image by comparing the image with its reference image, while NR-IQA evaluates an image without involving reference images. Meanwhile, RR-IQA assesses an image using partial information from reference images [12]. Notably, NR-IQA is the most suitable metric used to assess wood images due to the impediments (dusty environment and poor illumination) to the achievement of high-quality images in the environment of lumber mills.

Several NR-IQA metrics were proposed, such as Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) [13], deepIQA [12] and Deep Bilinear Convolutional Neural Networks (DB-CNN) [14]. Specifically, BRISQUE [13] is trained with Generalised Gaussian Distribution (GGD) and Asymmetric Generalised Gaussian Distribution (AGGD) features by Support Vector Regression (SVR) model for modelling of the images in the spatial domain. Furthermore, deepIQA and DB-CNN are CNN-based NR-IQAs, in which the deepIQA is trained end-to-end. It also involves 10 convolutional layers, five pooling layers for feature extraction, and two fully connected layers for regression [12]. Meanwhile, DB-CNN is trained by two sets of features, namely CNN for synthetic distortions (S-CNN) and VGG-16, which are bi-linearly pooled to measure the quality of the image [14]. However, provided that a limited number of labelled training data often leads to overfitting problem in CNN, the CNN-based NR-IQA model requires a larger size of the training database [14].

Accordingly, an investigation was conducted on the NR-IQA procedure, which was based on a widely-used NR-IQA, the Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) model. As an IQA model, BRISQUE was not a distortion-specific model. Instead, it considered the luminance and image features of the natural images [13]. Furthermore, the BRISQUE model was trained with subjective scores to enable the emulation of human judgement on the quality of the images. Provided that BRISQUE was trained to evaluate natural images, it was not optimal for the assessment of wood images. Therefore, an NR-IQA was proposed specifically for the assessment of wood images. Following that, the proposed metric, Wood NR-IQA (WNR-IQA) was then compared with BRISQUE [13], deepIQA [12], DB-CNN [14], and five types of established FR-IQA metrics, such as Structural Similarity Index (SSIM) [15], Multiscale SSIM (MS-SSIM) [15], Feature Similarity (FSIM) [16], Information Weighted SSIM (IW-SSIM) [17], and Gradient Magnitude Similarity Deviation (GMSD) [18]. The relative performances of the WNR-IQA, BRISQUE, deepIQA, DB-CNN, and FR-IQAs were identified based on the correlation between the human mean opinion scores (MOS) and the metrics. In this case, the Pearson Linear Correlation Coefficient (PLCC) [19] and Root Mean Squared Error (RMSE) [20] were used.

Materials and methods

Training and testing database

To cater to image quality assessment, specifically for wood images, Generalized Gaussian Distribution (GGD) and Asymmetric Generalized Gaussian Distribution (AGGD) features were calculated for wood images. This process involved the subjective MOS obtained from a subjective evaluation for wood images. These GGD, AGGD features, and MOS were used as the training and testing database for the SVR model.

Wood images

Ten wood images of ten wood species in the lumber industry, namely Turraeanthus africanus (Avodire), Ochroma pyramidale (Balsa), Cordia spp. (Bocote), Juglans cinerea (Butternut), Tilia Americana (Basswood), Vouacapoua americana (Brownheart), Cornus florida (Dogwood), Cordia spp. (Laurel Blanco), Swartzia Cubensis (Katalox), and Dipterocarpus spp (Keruing), were obtained from a public wood database: https://www.wood-database.com/ [21]. The ten wood images are presented in Fig 1.

Fig 1

Ten reference wood images (a) Turraeanthus africanus, (b) Ochroma pyramidale, (c) Tilia americana, (d) Cordia spp., (e) Juglans cinerea, (f) Vouacapoua americana, (g) Dipterocarpus spp., (h) Swartzia Cubensis, (i) Cordia spp., (j) Cornus florida.

Reprinted from [21] under a CC BY license, with permission from Eric Meier, original copyright [2007].

The images were then converted to grayscale, followed by normalisation of the pixel values to the range of 0–255 to facilitate the application of the same levels of distortion across all the reference images. Furthermore, the images consisted of a matrix of 600 x 600 pixels, which corresponded of an image area of 9525 cm2. The ten reference wood images were then distorted by Gaussian white noise and motion blur to represent the image distortions, which were encountered in the industrial setting. To be specific, Gaussian white noise often arises during the acquisition of wood images due to the sensor noise [22] caused by poor illumination and high ambient temperature in the lumber mill [8]. Meanwhile, wood images were subjected to motion blur upon the presence of relative motion between the camera and the wood slice [6]. Provided that these distortions resulted in a low quality of the wood image, the features of the pores on the wood texture could not be distinguished from one another. As a result, misclassification of the wood species occurred as the feature extractor would not be able to effectively extract distinctive features from the wood texture images [23].

The Gaussian white noise with a standard deviation of σ and a motion blur with a standard deviation of σ were applied to the reference images at five levels of distortion. For example, the σ for Gaussian white noise amounted to 10, 20, 30, 40, 50 and, while the σ for motion blur amounted to 2, 4, 6, 8 and 10. As a result, 110 wood images, ten reference images, 50 images distorted by Gaussian white noise, and 50 images distorted by motion blur were produced. This was followed by the measurement of GGD and AGGD features for these images, which were then used to train the SVR.

The features of GGD and AGGD

The Mean Subtracted Contrast Normalized (MSCN), was calculated using (1) [12]:

Where, I(m,n) denotes an image, while μ(m,n) denotes the local mean of I(m,n). While σ(m, n) refers to the local variance of I(m,n), m ∈ 1, 2, …, M, n ∈ 1, 2, …, N refers to the spatial indices, while M and N represent the height and width of image, I(m,n), respectively.

The local mean, μ(m,n), and local variance, σ(m,n), were calculated using the equations in (2) and (3), respectively [12]:

Where, w = {w|k = −K, …, K, l = −L, …, L} denotes a 2-dimension (2D) circularly-symmetric Gaussian weighting function, which was sampled in three standard deviations. This function was then rescaled to unit volume, in which K and L represent the window sizes. It could be seen in Fig 2 that the MSCN, local mean, and local variance of the wood images illustrate the effects of the contrast normalisation.

Fig 2

The effects of the image normalisation procedure on wood image.

Results are focused on the representative case of to Swartzia Cubensis: (a) Original image I, (b) Local mean-field, μ, (c) I − μ, (d) Local variance field, σ and (e) MSCN coefficients. Reprinted from [21] under a CC BY license, with permission from Eric Meier, original copyright [2007].

Furthermore, it is also indicated from Fig 2d that the local variance field, σ, only highlighted the boundary of the pores, while Fig 2e indicates that the MSCN coefficients focused on the key elements of the wood images, including pores and grains, with few low-energy residual object boundaries. According to Mittal et al., the characteristics of MSCN coefficients vary with the occurrence of the distortions [12]. Therefore, the MSCN coefficients were plotted for the reference image, including the images distorted with Gaussian white noise and motion blur to illustrate the resultant changes in the coefficients, as shown in Fig 3.

Fig 3

Histogram of MSCN coefficients for the reference image and distorted images with Gaussian white noise (GWN) and motion blur (MB).

Reprinted from [21] under a CC BY license, with permission from Eric Meier, original copyright [2007].

Based on Fig 3, a Gaussian distribution is presented in the reference images, while the distribution of the images distorted with Gaussian white noise and motion blur consisted of different tail behaviours.

Two types of Gaussian distribution functions were incorporated in this study to accommodate the diverse characteristics of MSCN coefficient, namely the Generalized Gaussian Distribution (GGD) and Asymmetric Generalized Gaussian Distribution (AGGD) [12].

There are two parameters computed for the GGD, where α represents the shape of the distribution and σ2 represents the variance. These two parameters are calculated for wood images using the moment-matching principle. The GGD is computed using (4) [24]: Where

Next, the MSCN coefficients were computed throughout the eight orientations, namely horizontal (H1 and H2), vertical (V1 and V2), and diagonal (D1, D2, D3 and D4) as shown in Fig 4.

Fig 4

Eight orientations of neighbourhood pixels of wood images: Horizontal (H1 and H2), vertical (V1 and V2) and diagonal (D1, D2, D3 and D4).

Reprinted from [21] under a CC BY license, with permission from Eric Meier, original copyright [2007].

The computation of the pairwise products of MSCN coefficients throughout the eight orientations: H1, H2, V1, V2, D1, D2, D3 and D4 are shown from Eqs (8) to (15) [12]:

Where, m ∈ 1, 2, …, M, n ∈ 1, 2, …, N, while M and N represent the height and width of the image.

The histogram of the pairwise products of MSCN coefficients throughout the eight orientations is presented in Fig 5.

Fig 5

Histogram of pairwise products of MSCN coefficients in eight directions: (a) D1 (b) D2 (c) D3 (d) D4 (e) H1 (f) H2 (g) V1 (h) V2 for the reference image and images distorted with Gaussian white noise (GWN) and motion blur (MB).

Reprinted from [21] under a CC BY license, with permission from Eric Meier, original copyright [2007].

The difference between pairwise products of MSCN coefficients along H1 and H2, V1 and V2, D1 and D3, and D2 and D4 were calculated, which indicates that H1 = H2, V1 = V2, D1 = D3, and D2 = D4. Hence, four orientations, namely H1, V1, D1 and D2 were chosen for the AGGD calculations.

Four parameters were computed for AGGD, namely η, v, and . Specifically, υ represents the shape of the distribution, and represent the left- and right-scale parameters, and η represents the mean of the distribution. The four parameters of AGGD, namely η, v, , , were calculated using the formula in (16) [25]. The AGGD parameters, η, v, were calculated throughout H1, V1, D1 and D2 orientations as shown in Eq (17) and this forms 16 parameters of AGGD. Where:

The parameters of the best AGGD fit were computed using the similar moment-matching approach, which was used for GGD, while η was calculated using the formula in (20) [12]:

In total, calculations were performed on 18 parameters of GGD and AGGD for the wood images, such as two parameters of GGD: α, σ2, 16 parameters of AGGD: four AGGD parameters, η, v, x 4 orientations, including H1, V1, D1, and D2, as shown in Table 1.

Table 1

Explanation of 18 parameters.

Features	Description	Computation procedures
f1 − f2	α and σ2	GGD features
f3 − f6	υ, η, σl2 and σr2	Horizontal (H1) AGGD features
f7 − f10	υ, η, σl2 and σr2	Vertical (V1) AGGD features
f11 − f14	υ, η, σl2 and σr2	Diagonal (D1) AGGD features
f15 − f18	υ, η, σl2 and σr2	Diagonal (D2) AGGD features

According to Mittal et al., accurate assessment of images could be conducted through IQA, which presents multi-scale information of an image [12]. Therefore, the aforementioned 18 parameters were computed at two scales (original image scale and image reduced by a factor of 0.5). Therefore, 36 parameters were generated from the full procedure to represent the features of wood images, and all parameters were used to train the SVR. As a result, only two scales were used, which reflected Mittal et al.’s statement that no improvement took place in the performance of the metric when more scales were incorporated [12]. The computation time would also increase with the increasing number of scales.

MOS

Ten students from the Department of Electrical and Electronics Engineering in Manipal International University (MIU), Nilai, Malaysia, who aged 20 to 25 years old, volunteered to evaluate the wood images. The evaluation was performed using a 21 inch LED monitor with a resolution of 1920 x 1080 pixels based on the procedures recommended in Rec. ITU-R BT.500-11 [26] within an office environment. The uncorrected near vision acuity of every subject was checked using the Snellen Chart prior to the subjective evaluation to confirm their fitness to perform the evaluation task.

After the examination of the uncorrected near vision acuity, a subjective evaluation was conducted. Consisting of a process which took 15 to 20 minutes, the evaluation was performed based on the Simultaneous Double Stimulus for Continuous Evaluation (SDSCE) methodology [26, 27]. In this case, the reference and distorted images were displayed on the monitor screen side-by-side, where the reference image was displayed on the left and the distorted image was displayed on the right. The distorted image was evaluated by each subject through the comparison between the distorted image (right side) and reference image (left side) in terms of quality. The image was either rated as Excellent (5), Good (4), Fair (3), Poor (2), or Bad (1) based on each displayed image. However, the numerical scores were not revealed to the subjects due to the potential bias created between the subjects [25]. The ratings obtained from the subjects were used to calculate MOS using the formula in (21) [28]:

Where S refers to the score by i subject for p image, while N represents the number of human subjects as N = 10. The MOS values obtained for wood images were also used to train SVR.

Regression module

An epsilon-SVR, ∈ − SVR model was used in this study [29]. As previously mentioned, the ∈ − SVR was trained using MOS, 36 GGD, and AGGD features of wood images. Following the calculation of 36 image features for the wood images, mapping of the features to MOS values of the respective wood images were performed. The 36 features and MOS of wood images were then divided randomly into two sets, where one set was used for training and another set was used to test the system. While 80% of the 36 features and MOS values were used to train the SVR model, the remaining 20% were used to test the system. The training and testing datasets were permutated randomly to avoid any bias during the training and testing of the system [12].

The difference between BRISQUE and WNR-IQA could be seen from how BRISQUE is the generalised form of IQA, which is made to obtain quality score for natural images, while WNR-IQA is created specifically for wood images. Natural images are any natural light images which are captured by an optical camera without any pre-processing [12]. While, wood images are captured using a portable camera which has ten times magnification lens [3]. The differences between BRISQUE and WNR-IQA flowcharts are presented in Fig 6.

Fig 6

Differences between BRISQUE and WNR-IQA.

Pearson’s Linear Correlation Coefficient (PLCC) [19] and Root Mean Square Error (RMSE) [20] between the MOS values and the quality score, which were obtained from the WNR-IQA, were calculated to evaluate the performance of the system. The accuracy of the system was indicated through higher PLCC and lower RMSE values due to the high similarity between the quality scores obtained from the WNR-IQA to the MOS values in terms of magnitude. The training and testing of the system were iterated 100 times, while the PLCC and RMSE values were recorded for every iteration. As a result, the medians of PLCC and RMSE amounted to 0.935 and 0.361, respectively. Moreover, the optimised cost parameter (C) and width parameter (g) of the SVR model, which amounted to 512 and 0.25, respectively, were selected based on the median of the PLCC and RMSE values. Following that, these parameters were used to form the optimised SVR model.

Performance evaluation

The second dataset was created specifically to evaluate the performance of WNR-IQA, where only the second dataset was used instead of the wood images highlighted in the Wood images sub-section for the evaluation of the performance of WNR-IQA. To illustrate, provided that the wood images were used for the training of the SVR, the second dataset was created to avoid any bias in performance evaluation. This dataset was generated using 10 ‘perfect’ reference images obtained from ten different wood species, namely Julbernardia pellegriniana (Beli), Dalbergia cultrate (Blackwood), Dalbergia retusa (Cocobolo), Dalbergia cearensis (Kingwood), Guaiacum officinale (Lignum), Swartzia spp. (Queenwood), Dalbergia spruceana (Rosewood), Dalbergia sissoo (Sisso), Swartzia benthamiana (Wamara), and Euxylophora paraensis (Yellowheart). These images are presented in Fig 7.

Fig 7

Ten reference wood images in the second dataset (a) Julbernardia pellegriniana, (b) Dalbergia cultrate, (c) Dalbergia retusa, (d) Dalbergia cearensis, (e) Guaiacum officinale, (f) Swartzia spp., (g) Dalbergia spruceana, (h) Dalbergia sissoo, (i) Swartzia benthamiana, and (j) Euxylophora paraensis.

Reprinted from [21] under a CC BY license, with permission from Eric Meier, original copyright [2007].

These images were obtained from the same wood image database [21] and distorted with Gaussian white noise with σ = 10, 20, 30, 40, and 50, including a motion blur with σ = 2, 4, 6, 8, and 10. Using motion blur, further distortion was performed on the images distorted by the Gaussian white noise. In this case, following the distortion of images with σ = 10 was further distortion with σ = 2, 4, 6, 8 and 10, and these procedures were repeated for images distorted with σ = 20, 30, 40, and 50. Overall, 360 wood images were generated in the dataset.

The proposed WNR-IQA metric was compared with five FR-IQA metrics obtained for the second dataset, namely Structural Similarity Index (SSIM) [15], Multiscale SSIM (MS-SSIM) [15], Feature Similarity (FSIM) [16], Information Weighted SSIM (IW-SSIM) [17], and Gradient Magnitude Similarity Deviation (GMSD) [18]. In addition, the WNR-IQA was also compared with BRISQUE [13], deepIQA [12], and DB-CNN [14] obtained for the second dataset. This was followed by the calculation of PLCC and RMSE [19] values between the FR-IQAs, BRISQUE, deepIQA, DB-CNN, and WNR-IQA for the evaluation of the performance of the WNR-IQA, BRISQUE, deepIQA, DB-CNN, and FR-IQAs.

Results and discussions

The relationship between MOS and different distortion levels

The relationship between MOS and different distortion levels of Gaussian white noise, motion blur, and a mixture of Gaussian white noise and motion blur is presented from Fig 8a–8g. Higher MOS values indicated higher image quality, while higher distortion levels represented lower image quality. Therefore, lower MOS values would be produced for images with higher distortion levels. Based on the scatter plot presented in Fig 8a–8e, the MOS value was reduced with the increase in distortion level. Accordingly, it was indicated that human subjects were able to differentiate the images distorted with different levels of Gaussian white noise, motion blur, and the mixture of both distortions. It could be seen from the scatter plot in Fig 8f and 8g that the MOS value amounted to 1, while the images distorted with Gaussian white noise, σ, amounted to 40 and 50 at all the levels of motion blur due to the poor quality of the images.

Fig 8

Scatter plot of MOS versus distortion levels of (a) Gaussian white noise, (b) Motion blur, (c) Motion blur for Gaussian white noise, σ = 10, (d) Motion blur for Gaussian white noise, σ = 20, (e) Motion blur for Gaussian white noise, σ = 30, (f) Motion blur for Gaussian white noise, σ = 40 and (g) Motion blur for Gaussian white noise, σ = 50.

Reprinted from [21] under a CC BY license, with permission from Eric Meier, original copyright [2007].

Relationship between MOS and proposed WNR-IQA, BRISQUE, deepIQA, DB-CNN, FR-IQAs

The calculated PLCC and RMSE values between MOS and the WNR-IQA, BRISQUE, deepIQA, DB-CNN, and the five FR-IQA metrics are presented in Table 2. PLCC values close to 1, indicate a close correlation of MOS with the IQA metric, while lower RMSE values indicate a correlation of MOS with the IQA metric. Table 2 shows that the highest PLCC values were recorded for Gaussian white noise, motion blur, the mixture of Gaussian white noise and motion blur, and the overall database obtained for the WNR-IQA compared to BRISQUE, deepIQA, DB-CNN, and five FR-IQAs. Therefore, WNR-IQA displayed a higher performance compared to BRISQUE, deepIQA, DB-CNN, SSIM, MS-SSIM, FSIM, IW-SSIM, and GMSD.

Table 2

PLCC and RMSE values between MOS and WNR-IQA, BRISQUE, deepIQA, DB-CNN, and FR-IQAs.

		WNR-IQA	BRISQUE	deepIQA	DB-CNN	MSSIM	SSIM	FSIM	IWSSIM	GMSD
PLCC	GWN	0.928	0.592	0.675	0.759	0.857	0.871	0.898	0.865	0.894
	MB	0.960	0.586	0.698	0.725	0.863	0.818	0.920	0.914	0.865
	Mixture of GWN and MB	0.925	0.591	0.642	0.698	0.861	0.832	0.895	0.880	0.872
	All	0.943	0.612	0.639	0.727	0.850	0.815	0.903	0.876	0.895
RMSE	GWN	0.481	1.039	0.945	0.867	0.664	0.633	0.568	0.647	0.578
	MB	0.321	0.925	0.923	0.910	0.577	0.656	0.448	0.462	0.572
	Mixture of GWN and MB	0.352	0.930	0.929	0.873	0.624	0.645	0.529	0.551	0.574
	All	0.385	0.916	0.911	0.890	0.623	0.721	0.456	0.529	0.539

It is also indicated from Table 2 that the lowest PLCC values were recorded for BRISQUE, indicating that BRISQUE was not compatible with the assessment of wood images. This incompatibility was also indicated by the highest RMSE values recorded for BRISQUE. However, WNR-IQA had a higher performance compared to BRISQUE, deepIQA, DB-CNN, and FR-IQAs as it was adapted for wood images. The model was also trained with GGD and AGGD features, including the MOS obtained for wood images unlike BRISQUE, deepIQA, DB-CNN, and FR-IQAs, which were designed based on the features and their similarities, luminance, contrast, and structure of natural images. Additionally, WNR-IQA also had a higher performance compared to FR-IQAs as it does not require a perfect reference image.

Conclusion

In this article, Wood No-Reference Image Quality Assessment (WNR-IQA), was proposed for the evaluation of wood images prior to classification of species. Provided that the established NR-IQA metrics, BRISQUE, deepIQA and DB-CNN were designed for the assessment of natural images, they were not optimal for the assessment of wood images. Therefore, the WNR-IQA was trained using MOS and a set of features calculated specifically for wood images. This was followed by the evaluation of the performance of the WNR-IQA by comparing the correlation between MOS, WNR-IQA, BRISQUE, deepIQA, DB-CNN, and five FR-IQA metrics using PLCC and RMSE. It was indicated from the values of PLCC and RMSE that WNR-IQA exhibited higher performance compared to BRISQUE, deepIQA, DB-CNN, and the five FR-IQAs. Furthermore, the proposed WNR-IQA performed an accurate assessment of the quality of wood images, which should function in the selection of suitable images to be included in the wood recognition algorithm. Essentially, the acquirement of a perfect image is impossible in the timber industry due to its environment, which consists of dust, poor illumination, hot environment, and motion blur caused by relative motion between the camera and the wood slice. However, the quality assessment in this study did not require a perfect reference image for the evaluation of the quality of the test wood images.

20 Jan 2020

PONE-D-19-34649

No-Reference Quality Assessment for Image-Based Assessment of Economically Important Tropical Woods

PLOS ONE

Dear Dr Mokhtar,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

We would appreciate receiving your revised manuscript by Mar 05 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Yiming Tang, Ph.D.

Academic Editor

PLOS ONE

Journal Requirements:

1. When submitting your revision, we need you to address these additional requirements.

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at

http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

2. Please amend your Data availability statement to provide links/URLs to how other researchers may access the images and data used in this study.

3. We suggest you thoroughly copyedit your manuscript for language usage, spelling, and grammar. If you do not know anyone who can help you do this, you may wish to consider employing a professional scientific editing service.

Whilst you may use any professional scientific editing service of your choice, PLOS has partnered with both American Journal Experts (AJE) and Editage to provide discounted services to PLOS authors. Both organizations have experience helping authors meet PLOS guidelines and can provide language editing, translation, manuscript formatting, and figure formatting to ensure your manuscript meets our submission guidelines. To take advantage of our partnership with AJE, visit the AJE website (http://learn.aje.com/plos/) for a 15% discount off AJE services. To take advantage of our partnership with Editage, visit the Editage website (www.editage.com) and enter referral code PLOSEDIT for a 15% discount off Editage services.  If the PLOS editorial team finds any language issues in text that either AJE or Editage has edited, the service provider will re-edit the text for free.

Upon resubmission, please provide the following:

The name of the colleague or the details of the professional service that edited your manuscript

A copy of your manuscript showing your changes by either highlighting them or using track changes (uploaded as a *supporting information* file)

A clean copy of the edited manuscript (uploaded as the new *manuscript* file)

4. We note that Figures 1-8 in your submission contain copyrighted images. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution. For more information, see our copyright guidelines: http://journals.plos.org/plosone/s/licenses-and-copyright.

We require you to either (1) present written permission from the copyright holder to publish these figures specifically under the CC BY 4.0 license, or (2) remove the figures from your submission:

1.         You may seek permission from the original copyright holder of Figures 1-8 to publish the content specifically under the CC BY 4.0 license.

We recommend that you contact the original copyright holder with the Content Permission Form (http://journals.plos.org/plosone/s/file?id=7c09/content-permission-form.pdf) and the following text:

“I request permission for the open-access journal PLOS ONE to publish XXX under the Creative Commons Attribution License (CCAL) CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). Please be aware that this license allows unrestricted use and distribution, even commercially, by third parties. Please reply and provide explicit written permission to publish XXX under a CC BY license and complete the attached form.”

Please upload the completed Content Permission Form or other proof of granted permissions as an "Other" file with your submission.

In the figure caption of the copyrighted figure, please include the following text: “Reprinted from [ref] under a CC BY license, with permission from [name of publisher], original copyright [original copyright year].”

2.    If you are unable to obtain permission from the original copyright holder to publish these figures under the CC BY 4.0 license or if the copyright holder’s requirements are incompatible with the CC BY 4.0 license, please either i) remove the figure or ii) supply a replacement figure that complies with the CC BY 4.0 license. Please check copyright information on all replacement figures and update the figure caption with source information. If applicable, please specify in the figure caption text when a figure is similar but not identical to the original image and is therefore for illustrative purposes only.

Additional Editor Comments:

The work was not convincing in the current version. Two reviewers pointed out some tough comments, which required further in-depth study and response by the authors. It was demanded to provide the comparison with related top algorithms (mentioned by Reviewer 2).

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Partly

Reviewer #2: No

**********

2. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: No

Reviewer #2: No

**********

3. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: No

Reviewer #2: Yes

**********

5. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: In this paper, the authors proposed a No-Reference IQA (NR-IQA) metric to assess the quality of wood images, Wood NR-IQA (WNR-IQA). Support Vector Machine (SVM) Regression (SVR) was trained using Generalized Gaussian Distribution (GGD) and Asymmetric Generalized Gaussian Distribution (AGGD) features calculated for wood images and the mean opinion score (MOS) obtained from subjective evaluation. Compared to other six metric, namely Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) and five Full Reference-IQA (FR-IQA) metrics known as MSSIM, SSIM, FSIM, IWSSIM and GMSD, the proposed WNR-IQA had better performance in PLCC and RMSE values. Furthermore, it is meaningful that the proposed metric does not require a “perfect” reference image in order to evaluate the images.

But, this manuscript in this current form has some problems as follows:

1) The introduction is not enough because the references for the current researches are too few while the related analyses are not profound.

2) The metric most relevant to the proposed metric is the BRISQUE. But the authors did not show the differences between them, especially the difference in image features adoption and the difference in procedure of these two metrics. Please add relevant instructions to highlight the innovation of this paper.

3) In Eq. (4) and (7), what does the “x” mean?

4) In section “Results and Discussions”, the authors have only considered the Gaussian white noise and motion blur. Why the mixture of these two noises were ignored in simulating the image distortions encountered in the industrial setting?

5) In section “Results and Discussions”, why the ten wood images mentioned in “wood images” , namely Avodire, Bocote, Butternut, Basswood, Dogwood, Laurel Blanco, Katalox, and Keruing are not taken in to experiments? They are replaced by other ten species, why?

6) In section “Relationship between MOS and Proposed WNR-IQA, BRISQUE,FR-IQAs”, there is no detailed explanation about Fig 8. In addition, pictures shown in Fig 8a-u are blurred. It is unacceptable.

7) Please clearly explain the reasons in section “Results and Discussions” based on the expremental results, why the proposed metric WNR-IQA is superior to other six metrics? And why other related metrics have lower performance?

8) The 18 parameters of GGD and AGGD for wood images are keys for proposed WNR-IQA. But there are no equations or detailed instructions to show how to calculate these parameters in this paper. That should not be neglected or omitted.

9) English expressions need to be carefully checked out. For example, in the first paragraph of Introduction, there exist the following sentence (noting “it high strength and density”):

“Rosewood (Dalbergia sp.) is another expensive wood, sought after for instrument making and flooring due to it high strength and density.”

Besides, at line 111 to 112, there is a sentence (noting “were the distorted”):

“These ten reference wood images were the distorted by Gaussian white noise and motion blur, which represent image distortions typically encountered in the industrial setting.”

Totally speaking, this manuscript is not well-prepared.

Reviewer #2: The authors presented a No-Reference IQA metric to assess the quality of wood images. Some experiments are done and compared with some other metrics. The results are rather OK. However, I can not find the innovative points in the paper since the authors just utilize the regression module to assess the quality of wood images. Furthermore, many state-of-the-art references are not cited. The experiments just compared with traditional methods, and many recent methods, especial some deep learning based methods such as Hallucinated-IQA and deepIQA, are not used to compare.

Minor points:

What’s mean of \\in -SVR in Line 239?

In equ. 19, it’s better to add subscript k in MOS since you calculate the MOS of the k^{th} image.

**********

6. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

4 Mar 2020

5 March 2020

Dear Editor and Reviewers,

Thank you very much for considering our manuscript entitled “No-reference quality assessment for image-based assessment of economically important tropical woods”. We really appreciate your valuable suggestions and comments which help us to enhance the quality of our paper. We have revised the manuscript accordingly, with details below:

Reviewers' comments:

When submitting your revision, we need you to address these additional requirements.

Please ensure that your manuscript meets PLOS ONE's style requirements, including those for file naming. The PLOS ONE style templates can be found at http://www.journals.plos.org/plosone/s/file?id=wjVg/PLOSOne_formatting_sample_main_body.pdf and

http://www.journals.plos.org/plosone/s/file?id=ba62/PLOSOne_formatting_sample_title_authors_affiliations.pdf

The manuscript has been revised according to the PLOS ONE's style.

Please amend your Data availability statement to provide links/URLs to how other researchers may access the images and data used in this study.

The images and data used in this study are publicly available in https://github.com/Heshalini/Wood-Image-Quality/tree/b03346da9d1efb887ec6d390f235822cc9470f49.

We suggest you thoroughly copyedit your manuscript for language usage, spelling, and grammar. If you do not know anyone who can help you do this, you may wish to consider employing a professional scientific editing service.

The manuscript has been proofread and edited by professional proofreading and editing service, Proofreading by A UK PhD.

We note that Figures 1-8 in your submission contain copyrighted images. All PLOS content is published under the Creative Commons Attribution License (CC BY 4.0), which means that the manuscript, images, and Supporting Information files will be freely available online, and any third party is permitted to access, download, copy, distribute, and use these materials in any way, even commercially, with proper attribution.

Figs 1-8 were generated using the publicly available wood images database owned by Mr Eric Meier. Permission to publish the figures has been obtained from Mr Eric Meier and the completed Content Permission Form and e-mail from Mr Eric which mentioned that he permits the usage of the wood images from his website are uploaded as an "Other" file with this submission.

Reviewer 1:

In this paper, the authors proposed a no-reference IQA (NR-IQA) metric to assess the quality of wood images, Wood NR-IQA (WNR-IQA). Support Vector Machine (SVM) Regression (SVR) was trained using Generalized Gaussian Distribution (GGD) and Asymmetric Generalized Gaussian Distribution (AGGD) features calculated for wood images and the mean opinion score (MOS) obtained from the subjective evaluation. Compared to other six metrics, namely Blind/Referenceless Image Spatial Quality Evaluator (BRISQUE) and five Full Reference-IQA (FR-IQA) metrics known as MSSIM, SSIM, FSIM, IWSSIM and GMSD, the proposed WNR-IQA had better performance in PLCC and RMSE values. Furthermore, it is meaningful that the proposed metric does not require a “perfect” reference image in order to evaluate the images.

But, this manuscript in this current form has some problems as follows:

The introduction is not enough because the references for the current researches are too few while the related analyses are not profound.

Explanation on current researches done on CNN based NR-IQAs, deepIQA [12] and Deep Bilinear Convolutional Neural Networks (DB-CNN) [14] were added in line 81-106 of the Introduction section. DeepIQA and DB-CNN are CNN based NR-IQAs where deepIQA is trained end-to-end and involves 10 convolutional layers, 5 pooling layers for feature extraction and 2 fully connected layers for regression [12] while DB-CNN is trained by two sets of features namely, CNN for synthetic distortions (S-CNN) and VGG-16, that are bi-linearly pooled to predict the quality of the image [14]. However, CNN based NR-IQA model requires a very large training database as a limited number of labelled training data often leads to overfitting problem in CNN [14].

The metric most relevant to the proposed metric is the BRISQUE. But the authors did not show the differences between them, especially the difference in image features adoption and the difference in the procedure of these two metrics. Please add relevant instructions to highlight the innovation of this paper.

The innovation of the proposed WNR-IQA model is that it is a No Reference – Image Quality Assessment (NR-IQA) model designed specifically for wood images. WNR-IQA is motivated by NR-IQA model which is independent of reference natural images. In fact, for practicality, good reference images especially given the nature of the wood industry (dusty environment and poor illumination) are not easily available. These were written in line 78-80 of the Introduction section.

The proposed model uses a similar concept designed for natural images, with mathematical models that consider several factors like contrast, luminance, image feature, and Natural Scene Statistics (NSS). The key difference between the BRISQUE and WNR-IQA is that the BRISQUE was developed by training the SVR using the image features computed for natural images; whereas the WNR-IQA used the wood images for this calculation. These were written in line 277-282 of the Regression module section. Their differences were also illustrated in Fig 6, Regression module section.

In Eq. (4) and (7), what does the “x” mean?

Eq. (4) is the equation to calculate GGD features. x in Eq. (4) is the Mean Subtracted Contrast Normalized (MSCN), I ^(m,n) of wood images. This was written in Eq. (5), The features of GGD and AGGD section.

Eq. (7) (in previous manuscript version) is now Eq. (16) which is the equation to calculate AGGD features. x in Eq. (16) is the pairwise products of MSCN coefficients which are computed along four orientation, H_1,V_1,D_1 and D_2. This was written in Eq. (17), The features of GGD and AGGD section

In section “Results and Discussions”, the authors have only considered the Gaussian white noise and motion blur. Why the mixture of these two noises were ignored in simulating the image distortions encountered in the industrial setting?

All the ten wood images in the second dataset have been distorted with the mixture of Gaussian white noise and motion blur. The images which were distorted by Gaussian white noise were further distorted with motion blur, i.e. images distorted with σ_GN = 10 were further distorted with σ_MB = 2, 4, 6, 8 and 10 and the same procedure were repeated for images distorted with σ_GN = 20, 30, 40 and 50. In total, this dataset comprises of 360 wood images. These were written in line 314 – 317 of the Performance evaluation section. The results obtained for the mixture of Gaussian white noise and motion blur can be found in line 327– 365 of the Results and discussions section.

In section “Results and Discussions”, why the ten wood images mentioned in “wood images”, namely Avodire, Bocote, Butternut, Basswood, Dogwood, Laurel Blanco, Katalox, and Keruing are not taken into experiments? They are replaced by other ten species, why?

The ten wood images mentioned in the wood images section are used to train and test the SVR model. These images were distorted with Gaussian White Noise and Motion Blur and were used to train and test the SVR model for 100 times where 100 PLCC and RMSE values were obtained and the median of these values were chosen in order to select the optimized cost parameter, C, and width parameter, g, of the SVR model. These are written in line 269-293 of the Regression module section. The median values of PLCC and RMSE were also mentioned in these lines.

The performance of the proposed WNR-IQA metric is evaluated using the second dataset which also comprises of different ten wood images. Therefore, the results obtained for these images were included and discussed in the Results and Discussion section. The second dataset is used to evaluate the performance of WNR-IQA metric instead of the ten wood images mentioned in the wood images section (used to train SVR) to avoid any bias. These are written in line 295-299 of the Performance evaluation section.

In section “Relationship between MOS and Proposed WNR-IQA, BRISQUE, FR-IQAs”, there is no detailed explanation about Fig 8. In addition, pictures shown in Fig 8a-u are blurred. It is unacceptable.

Fig 8a-u which shows the distribution of the WNR-IQA, BRISQUE, FR-IQAs has been removed due to the poor quality of the images. Moreover, the relationship between MOS and Proposed WNR-IQA, BRISQUE, FR-IQAs is shown through the PLCC and RMSE values in Table 2, Relationship between MOS and proposed WNR-IQA, BRISQUE, deepIQA, DB-CNN, FR-IQAs section.

Please clearly explain the reasons in section “Results and Discussions” based on the experimental results, why the proposed metric WNR-IQA is superior to the other six metrics? And why other related metrics have lower performance?

The WNR-IQA was compared with BRISQUE, deepIQA, DB-CNN, five FR-IQAs: SSIM, MS-SSIM, FSIM, IW-SSIM and GMSD obtained for the second dataset. WNR-IQA outperforms BRISQUE, deepIQA, DB-CNN and FR-IQAs as it is tailored for wood images where the model has been trained with GGD and AGGD features and MOS obtained for wood images specifically unlike BRISQUE, deepIQA, DB-CNN and FR-IQAs which were designed by considering the features and similarities of features, luminance, contrast and structure of natural images. Furthermore, WNR-IQA is better than the FR-IQAs as it does not require a perfect reference image. These were written in line 347 – 365 of the Relationship between MOS and proposed WNR-IQA, BRISQUE, deepIQA, DB-CNN, FR-IQAs section.

The 18 parameters of GGD and AGGD for wood images are keys for proposed WNR-IQA. But there are no equations or detailed instructions to show how to calculate these parameters in this paper. That should not be neglected or omitted.

The equations and detailed instruction for GGD and AGGD parameters were added and can be found in line 182-234 and Table 1 of The features of GGD and AGGD section.

English expressions need to be carefully checked out. For example, in the first paragraph of Introduction, there exist the following sentence (noting “it high strength and density”):

“Rosewood (Dalbergia sp.) is another expensive wood, sought after for instrument making and flooring due to its high strength and density.”

Besides, at line 111 to 112, there is a sentence (noting “were the distorted”):

“These ten reference wood images were distorted by Gaussian white noise and motion blur, which represent image distortions typically encountered in the industrial setting.”

The above sentences were corrected and the manuscript has been proofread and edited by professional proofreading and editing service, Proofreading by A UK PhD.

Reviewer #2: The authors presented a no-reference IQA metric to assess the quality of wood images. Some experiments are done and compared with some other metrics. The results are rather OK. However, I can not find the innovative points in the paper since the authors just utilize the regression module to assess the quality of wood images. Furthermore, many state-of-the-art references are not cited. The experiments just compared with traditional methods, and many recent methods, especial some deep learning-based methods such as Hallucinated-IQA and deepIQA, are not used to compare.

The innovation of the proposed WNR-IQA model is that it is a No Reference – Image Quality Assessment (NR-IQA) model designed specifically for wood images. WNR-IQA is motivated by NR-IQA model which is independent of reference natural images. In fact, for practicality, good reference images especially given the nature of the wood industry (dusty environment and poor illumination) are not easily available. These were written in line 78 - 80 of the Introduction section.

The proposed model uses a similar concept designed for natural images, with mathematical models that consider several factors like contrast, luminance, image feature, and Natural Scene Statistics (NSS). The key difference between the BRISQUE and WNR-IQA is that the BRISQUE was developed by training the SVR using the image features computed for natural images; whereas the WNR-IQA used the wood images for this calculation. These were written in the line 277 - 282 of the Regression module section. Their differences were also illustrated in Fig 6.

The WNR-IQA was also compared with deepIQA and DB-CNN obtained for the second dataset. Results shows that WNR-IQA outperforms these two NR-IQAs as well. These were written in line 347 – 365 of the Relationship between MOS and proposed WNR-IQA, BRISQUE, deepIQA, DB-CNN, FR-IQAs section.

Minor points:

What’s mean of \\in -SVR in Line 239?

∈-SVR is epsilon-SVR which is a type of SVR model. This was written in line 269 of the Regression module section.

In equ. 19, it’s better to add subscript k in MOS since you calculate the MOS of the k^{th} image.

Subscript k was replaced with p to avoid confusion with the window sizes in Eq. (2) and (3). Subscript p was added in MOS equation in Eq. (21), MOS section.

Thank you for your kind consideration of our work.

Yours Faithfully,

NORRIMA MOKHTAR

Senior Lecturer

Department of Electrical Engineering

Faculty of Engineering

University of Malaya

50603 Kuala Lumpur

Malaysia

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

22 Apr 2020

PONE-D-19-34649R1

No-Reference Quality Assessment for Image-Based Assessment of Economically Important Tropical Woods

PLOS ONE

Dear Dr Mokhtar,

Thank you for submitting your manuscript to PLOS ONE. After careful consideration, we feel that it has merit but does not fully meet PLOS ONE’s publication criteria as it currently stands. Therefore, we invite you to submit a revised version of the manuscript that addresses the points raised during the review process.

We would appreciate receiving your revised manuscript by Jun 06 2020 11:59PM. When you are ready to submit your revision, log on to https://www.editorialmanager.com/pone/ and select the 'Submissions Needing Revision' folder to locate your manuscript file.

If you would like to make changes to your financial disclosure, please include your updated statement in your cover letter.

To enhance the reproducibility of your results, we recommend that if applicable you deposit your laboratory protocols in protocols.io, where a protocol can be assigned its own identifier (DOI) such that it can be cited independently in the future. For instructions see: http://journals.plos.org/plosone/s/submission-guidelines#loc-laboratory-protocols

Please include the following items when submitting your revised manuscript:

A rebuttal letter that responds to each point raised by the academic editor and reviewer(s). This letter should be uploaded as separate file and labeled 'Response to Reviewers'.

A marked-up copy of your manuscript that highlights changes made to the original version. This file should be uploaded as separate file and labeled 'Revised Manuscript with Track Changes'.

An unmarked version of your revised paper without tracked changes. This file should be uploaded as separate file and labeled 'Manuscript'.

Please note while forming your response, if your article is accepted, you may have the opportunity to make the peer review history publicly available. The record will include editor decision letters (with reviews) and your responses to reviewer comments. If eligible, we will contact you to opt in or out.

We look forward to receiving your revised manuscript.

Kind regards,

Yiming Tang, Ph.D.

Academic Editor

PLOS ONE

[Note: HTML markup is below. Please do not edit.]

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors have followed carefully our suggestions, and therefore I recommend accepting this paper for publication in PLOS ONE.

Reviewer #2: My comments are responded by authors carefully. The manuscript is improvement significantly. However, the model description is still not clear enough. Please give some explanation the difference between "natural image" and "wood image" . In Line 216, a parameter \\eta is missed

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: No

[NOTE: If reviewer comments were submitted as an attachment file, they will be attached to this email and accessible via the submission site. Please log into your account, locate the manuscript record, and check for the action link "View Attachments". If this link does not appear, there are no attachment files to be viewed.]

While revising your submission, please upload your figure files to the Preflight Analysis and Conversion Engine (PACE) digital diagnostic tool, https://pacev2.apexcovantage.com/. PACE helps ensure that figures meet PLOS requirements. To use PACE, you must first register as a user. Registration is free. Then, login and navigate to the UPLOAD tab, where you will find detailed instructions on how to use the tool. If you encounter any issues or have any questions when using PACE, please email us at figures@plos.org. Please note that Supporting Information files do not need this step.

28 Apr 2020

29 April 2020

Dear Editor and Reviewers,

Thank you very much for considering our manuscript entitled “No-reference quality assessment for image-based assessment of economically important tropical woods”. We really appreciate your valuable suggestions and comments which help us to enhance the quality of our paper. We have revised the manuscript accordingly, with details below:

Reviewers' comments:

Reviewer #2: My comments are responded by authors carefully. The manuscript is improvement significantly. However, the model description is still not clear enough. Please give some explanation the difference between "natural image" and "wood image" . In Line 216, a parameter \\eta is missed

The features of GGD and AGGD section was revised to make the model description clear. The content of the features of GGD and AGGD section was arranged to be in a flow to make the model description clear. Furthermore, line 227- 228 was revised by adding the following statement:

“The AGGD parameters, η,v,σ_l^2,σ_r^2 were calculated throughout H_1, V_1, D_1 and D_2 orientations as shown in equation (17) and this forms 16 parameters of AGGD.”

Table 1 which explains the 18 GGD and AGGD parameters was corrected as there was an error previously. The revised content can be found in line 146 – 250.

Fig 6 which shows the differences between BRISQUE and WNR-IQA was revised where the MSCN for the images was added into the flowchart to make the flowchart of the model clear.

The natural and wood images were explained in line 283-285, Regression module section. Natural images are any natural light images which are captured by an optical camera without any pre-processing. While, wood images are captured using a portable camera which has ten times magnification lens.

The missing parameter \\eta in line 216 has been revised. The revised line can be found in line 224, The features of GGD and AGGD section.

Thank you for your kind consideration of our work.

Yours Faithfully,

NORRIMA MOKHTAR

Senior Lecturer

Department of Electrical Engineering

Faculty of Engineering

University of Malaya

50603 Kuala Lumpur

Malaysia

Submitted filename: Response to Reviewers.docx

Click here for additional data file.

4 May 2020

No-reference quality assessment for image-based assessment of economically important tropical woods

PONE-D-19-34649R2

Dear Dr. Mokhtar,

We are pleased to inform you that your manuscript has been judged scientifically suitable for publication and will be formally accepted for publication once it complies with all outstanding technical requirements.

Within one week, you will receive an e-mail containing information on the amendments required prior to publication. When all required modifications have been addressed, you will receive a formal acceptance letter and your manuscript will proceed to our production department and be scheduled for publication.

Shortly after the formal acceptance letter is sent, an invoice for payment will follow. To ensure an efficient production and billing process, please log into Editorial Manager at https://www.editorialmanager.com/pone/, click the "Update My Information" link at the top of the page, and update your user information. If you have any billing related questions, please contact our Author Billing department directly at authorbilling@plos.org.

If your institution or institutions have a press office, please notify them about your upcoming paper to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, you must inform our press team as soon as possible and no later than 48 hours after receiving the formal acceptance. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information, please contact onepress@plos.org.

With kind regards,

Yiming Tang, Ph.D.

Academic Editor

PLOS ONE

Reviewers' comments:

Reviewer's Responses to Questions

Comments to the Author

1. If the authors have adequately addressed your comments raised in a previous round of review and you feel that this manuscript is now acceptable for publication, you may indicate that here to bypass the “Comments to the Author” section, enter your conflict of interest statement in the “Confidential to Editor” section, and submit your "Accept" recommendation.

Reviewer #1: All comments have been addressed

Reviewer #2: All comments have been addressed

**********

2. Is the manuscript technically sound, and do the data support the conclusions?

The manuscript must describe a technically sound piece of scientific research with data that supports the conclusions. Experiments must have been conducted rigorously, with appropriate controls, replication, and sample sizes. The conclusions must be drawn appropriately based on the data presented.

Reviewer #1: Yes

Reviewer #2: Yes

**********

3. Has the statistical analysis been performed appropriately and rigorously?

Reviewer #1: Yes

Reviewer #2: Yes

**********

4. Have the authors made all data underlying the findings in their manuscript fully available?

The PLOS Data policy requires authors to make all data underlying the findings described in their manuscript fully available without restriction, with rare exception (please refer to the Data Availability Statement in the manuscript PDF file). The data should be provided as part of the manuscript or its supporting information, or deposited to a public repository. For example, in addition to summary statistics, the data points behind means, medians and variance measures should be available. If there are restrictions on publicly sharing data—e.g. participant privacy or use of data from a third party—those must be specified.

Reviewer #1: Yes

Reviewer #2: Yes

**********

5. Is the manuscript presented in an intelligible fashion and written in standard English?

PLOS ONE does not copyedit accepted manuscripts, so the language in submitted articles must be clear, correct, and unambiguous. Any typographical or grammatical errors should be corrected at revision, so please note any specific errors here.

Reviewer #1: Yes

Reviewer #2: Yes

**********

6. Review Comments to the Author

Please use the space provided to explain your answers to the questions above. You may also include additional comments for the author, including concerns about dual publication, research ethics, or publication ethics. (Please upload your review as an attachment if it exceeds 20,000 characters)

Reviewer #1: The authors have followed carefully our suggestions and I recommend accepting this paper for publication in PLOS ONE.

Reviewer #2: All my comments are considered. I have no any further suggestion. The manuscript can be accepted now.

**********

7. PLOS authors have the option to publish the peer review history of their article (what does this mean?). If published, this will include your full peer review and any attached files.

If you choose “no”, your identity will remain anonymous but your review may still be made public.

Do you want your identity to be public for this peer review? For information about this choice, including consent withdrawal, please see our Privacy Policy.

Reviewer #1: No

Reviewer #2: Yes: Duanbing Chen

7 May 2020

PONE-D-19-34649R2

No-reference quality assessment for image-based assessment of economically important tropical woods

Dear Dr. Mokhtar:

I am pleased to inform you that your manuscript has been deemed suitable for publication in PLOS ONE. Congratulations! Your manuscript is now with our production department.

If your institution or institutions have a press office, please notify them about your upcoming paper at this point, to enable them to help maximize its impact. If they will be preparing press materials for this manuscript, please inform our press team within the next 48 hours. Your manuscript will remain under strict press embargo until 2 pm Eastern Time on the date of publication. For more information please contact onepress@plos.org.

For any other questions or concerns, please email plosone@plos.org.

Thank you for submitting your work to PLOS ONE.

With kind regards,

PLOS ONE Editorial Office Staff

on behalf of

Professor Yiming Tang

Academic Editor

PLOS ONE