Feature Based Stitching Of Liver Pathologic Biopsy Micorgraph

Image stitching plays an important role in digital image processing field, which used to solve the limitation of the viewfield. To make diagnosis of liver disease more accurately, the quantification analysis based on liver biopsy micrograph is often necessary. But because of the limitation of the microscope, the complete liver biopsy micrograph can not be taken by only one time, so the liver biopsy micrograph stitching is essential in quantification analysis.To get better accuracy and speed, a coarse-to-fine framework is used in the architecture. First we stitch the sampled image to get the coarse transforming parameter, then the parameter is refined in original image. To solve the image blurring between two adjoining images, we use scale-invariant feature detecting algorithm to get the interest point, and use orientation labeling method to make the feature descriptor rotation invariant, in the next step, we get the initial matching points set based on the similarity of the feature descriptor. Then use affine transforming as base model, use random sampling consensus to estimate the transforming parameter between two images, after get the approximate overlapping area between the two images, then we refining the parameter on the original image. Based on the parameter, the sequential image will be aligned and blended to get the final image with complete viewfield.Experiment on 150 cases (more than 1000 images) shows that the system is robust and adaptive, the result is good on the whole.