Research On Medical Image Feature Extraction And Recognition Algorithm Based On SIFT Feature And SVM Classifier

With the rapid development of medical visualization technology, Medical images play a more and more important role in clinical applications and medical research domain. The increase number of medical images makes how to manage those pictures and make use of it to be a new research hotspot.This thesis proposes a medical image recognition system based on SIFT feature extraction and SVM classification. Firstly, the article introduces the research status and significance about medical images recognition. By summarizing medical visualization technology and image feature theory, thesis gives an outline of different kind of medical images recognition systems. Taking the consideration medical image characteristics, thesis chooses to extract SIFT feature to build recognition system and describes the process of SIFT extraction in detail. Secondly, thesis uses K-mean to classify the image features. K is set a appropriate value to achieve the balance between accuracy of distinguish and cost of calculate. Then, being inspired by the thought of Bag of Image Features, thesis transforms medical images into statistical data of SIFT descriptors’classification. The data is given by the proportion of each sort of descriptors in total descriptors. After that, a medical image could be expressed by a k-dimensional vector. SVM classifier would use those vectors to distinguish one kind of images from others. As a classical classifier, SVM is designed to solve Two-class problem. But, In the practical application of medical images’ recognition, there always be a multi-class problem. So the thesis chooses lvl method in multi-class SVM to solve it.In the simulation, thesis classifies four kinds of medical images and analyses several parameters included in the system. The finally result of recognition accuracy is satisfaction. The whole system achieves the automation of medical image processing, could be used for medical image management and clinical applications.