Efficient large scale image stitching for virtual microscopy.

We describe a weighted least squares based global registration method for creating virtual slides. In microscopy a large number of fields of view is required to capture a complete slide. These fields of view are aligned in such a way that a globally consistent virtual slide is formed. The positioning accuracy of the presented algorithm is evaluated using a new method based on synthetic virtual slides. Using these synthetic virtual slides it is now possible to give a ground truth about the positioning accuracy of stitching algorithm for virtual microscopy. Our algorithm has been evaluated on these synthetic slides and it can be shown that on average, each image exhibits a mean deviation of 0.8 pixel with respect to the correct position. Additionally the presented algorithm has been evaluated on several real world examples.