Research On Automated Stitching Critical Technology Of Biomedical Image

Image stitching is used for generating a panoramic image with two or three pictures overlapped with each other. It has many practical applications, such as clinical diagnosis (scoliosis and lower limb malformation) and observing the biomedical sample with microscope. However, the available stitching methods have some defects such as the unacceptable deviation, low speed, and the need of manual intervention. Therefore, this paper mainly concentrates on studying the stitching of X-ray medical images and microscope images.For X-ray images, this paper proposed two kinds of seamless stitching methods based on contour feature and point feature. In the improved method based on contour feature, the image size is decreased to reduce the amount of computation and the accurate registration is accomplished by constructing the scale-space of the images, detecting the edge of the images, and utilizing the phase correlation repeatedly to identify the overlapping area. In the improved method based on point feature, pre-processing filter and phase correlation are employed to identify the rough overlapping area, where the extracted feature points are matched by the algorithm called mutual correlation of adjacent region of corner points. Finally, the linear blending method is employed to eliminate the fusion seam. The comparison of the results of our method with the clinical software demonstrates that the performance of the algorithm proposed in this paper can reach the level of the available clinical software.For microscope images, in order to overcome the defect of the SURF (ie., the number of the extracted features is small, so is the matched features), an image feature extraction algorithm, R-SURF, which is based on histogram equalization and scale-space reconstruction, is proposed to extract the features of microscope images. Then, the images are stitched automatically and seamlessly. Lastly, a novel affine invariant feature extraction algorithm (A-SURF) which is based on SURF was proposed by researching and analyzing the affine transformation of microscope images. A-SURF is robust against change of viewpoint, blur, illumination, rotation, and it can be applied for feature extraction and matching of images with great changes of viewpoint.