Research On Algorithm Of Medical Image Feature Description And Registration

Feature description and matching are an essential component of many computer vision applications. Feature description and image registration are the basis of image processing, analyzing and understanding, widely used in remote sensing image processing, medical image processing and other fields. Image feature description means quantitative analysis the region of interest of the image. Image registration technology solves the problem of alignment of several images. As the basis of image processing field, the results of algorithm affects the precision of the subsequent image processing directly.Benign and malignant melanoma can be effectively distinguished by irregular information of Medical tumor outline. The description of irregular information of tumor outline has important clinical significance to the diagnosis of melanoma. This paper introduces the description parameters of the image feature such as image outline, image texture, and the description method, as well as the application in the medical auxiliary diagnosis and treatment. This paper presents a description method of medical imaging tumor contour.Image registration is the premise of the image processing, and the key step of the image analysis. Image registration can be divided into image registration algorithm based on features and image registration based on pixels. The image registration algorithm based on feature transforms the whole image analysis to the analysis of image feature. This method can greatly reduce the computational complexity in the image processing. Therefore the paper introduces the existing image registration algorithms, and further studies the feature-based image registration method. How to extract the reliable and stable image features? How to improve the accuracy of the feature matching? These are important parts of the image registration. This paper proposes a feature-based eye fundus images registration method, which improved the complexity and the accuracy of the existing registration algorithm.The main contents of the paper can be summarized as follows:(1) This paper proposes a local structural features description of tumor outline based on the Gabor and fractal description method. Image texture feature can be effectively extracted based on Meyer’s cartoon-texture decomposition model. The best decomposition scale of the feature can be quickly and accurately extracted by Gabor scaling kernel. Also, the scale and the asymmetric information of tumor outline are described by the local scale and the local fractal dimension. Compared with the traditional description method, the novel algorithm has strong distinguish and high classification accuracy. The descriptions of medicine tumor Outlines has higher diagnostic significance.(2) An image registration algorithm based on Gabor filter and AP clustering is put forward. There are many factors that results in the mismatch of the image registration algorithm, Such as the number of feature points, the distinctiveness between feature points, the search strategy of matching, similarity measure, etc. Existing feature point extraction algorithm extracts amount of feature points which have limited distinctiveness. To solve the problem of mismatching caused by existing algorithm, improved algorithm is proposed, which extracts feature points according with human vision characteristic based on multi-scale analysis of Gabor filter. The128dimensions descriptor makes the feature has high uniqueness and distinctiveness. The improved algorithm reduces false match by improving the matching search strategies. It based on space and distance constraints on feature points through the AP clustering analysis, and Euclidean distance is taken as similarity measure. The simulation experiment results show that the algorithm can extract steady feature points accurately, and remove mismatching caused by similar content of image effectively.