A dual approach to structural texture analysis in microscopic cell images.

The computer-based quantitative analysis of microscopic cell images provides important diagnostic information in clinical and experimental pathology. The arrangement of various cell structures can be described as texture. We developed a new approach to structural texture analysis. It assumes that texture consists of homogeneous regions (texture primitives). Texture can be regarded in a dual way--as a composition of regions or as a pattern composed of the regions' boundaries (lines). We implemented methods for the segmentation of regions and lines in grayscale images. The detection of regions is followed by a region-growing process to avoid an oversegmentation. The segmented regions and lines are stored in a uniform data structure which reflects their arrangement in the image. The presented methods were applied to study the chromatin distribution in cell nuclei and the development and differentiation of intermediate filaments in fetal liver cells.