Research Of Image Classification Based On The Fusion Of SURF And Global Features

With fast progress of Network and medical image technology, medical images for teaching and clinical research are growing explosively in recent years. How to verdict image categories accurately and efficiently, and search out the similar medical image, which has become the emphasis of research in the medical image database currently. Medical images are often affected by medical imaging equipment, noises, fuzzy uneven, which has increased the difficulty of medical image classification, so getting stability features of images accurately is very important.This paper studies the local features and global features of images, and uses SVM (Support Vector Machine) to classify these two different features of images respectively, at last, the two classification results were integrated by the algorithm of decision-level integration to get the ultimate classification result. The main works of this paper are as follows:(1) This paper studied global features, local features and the algorithm of LSH (Locality Sensitive Hashing). The vector sets of SURF (Speed Up Robust Features) were reduced to a single high dimensions feature vector by employing the random histogram algorithm with LSH.(2) Based on the research of SVM, this paper proposes the decision level algorithm of multi-features based on SVM. Firstly, the global feature and the local feature are classified by using SVM respectively, and then, these two classification results are integrated by decision level algorithm to get the ultimate result, which significantly improves the classification accuracy(3) This paper adapts different features and different fusion algorithms to do classification experiments. Experimental results demonstrate that the decision level algorithm can improve the accuracy of medical image classification on the basis of effects of comparison and analysis of different features and different fusion strategies.(4) Under the Linux system, we used C++to develop the software of medical images classification based on the fusion of SURF and global features with computer vision library Opencv. This system has a user access interface which is similar to Google.and returns a series of similar images after user submits a medical image. The system uses the B/S architecture, deploys the Sever to the Linux platform and uses Apache as the http Server, and then develops the medical image classification system. Experimental results demonstrate that this paper’s algorithm can improve the accuracy of medical image classification.