Medical image mosaic based on low-overlapping regions

This paper focuses on the mosaicing of microscopic medical images. Different from regular application scene where the source images usually have an overlapping ratio between 30%∼50%, source images in this paper have a rather low overlapping ratio less than 15%. Performances of several generic feature-based registration methods are compared in order to explore which one is most appropriate for our application. This paper presents a novel mosaic algorithm as follows. Firstly, Speeded-up Robust Feature (SURF) points in the regions surrounding the overlapped are extracted and then roughly matching results of feature points are acquired. Secondly, mismatched pairs are eliminated according to spatial distance of matching points, and then geometrical relationship between reference image and target image is determined. Finally, registered images are blended into a composite image with modified weighted smooth algorithm. Experimental results demonstrate that the proposed method is effective to create a visually pleasant panoramic image.