| It is a key technology to diagnose and treatment glaucoma that automatically assessment anterior chamber angle timely in OCT images. Now, researchers measure and analyze the detailed data of all kinds of structural parameters mainly in term of the specific structure of anterior chamber angle, which leads to the facts that not only the measurement tasks are arduous, but also the evaluation is highly subjective. This paper tries to use of the image feature to describe eye anterior chamber Angle in OCT images, and then using the support vector machine (SVM) to evaluate the degree of anterior chamber Angle. First, interested front corner area is extracted with the feature edge, angle of key areas of the image is divided crypt in the paper, then describe the interested areas with features such as HOG, COHOG, and use SVM for image classification. Finally, appropriate characterization algorithm were considered in the point, line, surface three levels, and use multi-features to describe angle recess area, and classify the image to multi-levels with the Shaffer rating methodology.The main work and research results of this thesis are summarized as follows:(1) A method for extraction the region of interest based on edge features automatically was proposed. First, considering the various types of image noise and distortion, we proposed an algorithm for image segmentation with multi-level, which provided a premise for pinpoint the area of interest. Then, aimed at the characteristics of the front angle structure and the situation that there only the irisand the corneal region exist in the image, hence an edge detection algorithm was adopted which based on the direction of the continuity of pixel. for positioning the iris and the cornea edge quickly and easily. At last, we determined the intersection point of fitting a straight line through the edge liner fit and located the areas of interest according to the point of intersection.(2) This thesis puts forward a classification method for anterior chamber angle based on Region of Interest and HOG feature. The Region of Interest is located through the reference point calculated on the basis of image segmentation. The classifier training and image are finally classified by the combination of HOG feature and support vector machine. The test results show that firstly the region of interest automatic extraction methods proposed locate quickly and accurately, which provided a guarantee for feature extraction and the classification of samples, second, the combination of HOG feature and the support vector could distinguish closed and open state better, which made the classification better and stable.(3) This. thesis puts forward a classification method for OCT images of anterior chamber angle based on COHOG. While using the information of the pixels’ space position, I adopt a law-of-cosines method to calculate the weighed value, for the purpose of increasing the use of gradient magnitude and re-counting the COHOG feature.Then, COHOG characteristics of high-dimensional method of dimension reduction using PCA. At last, a SVM method classifying OCT images of anterior chamber into two groups is achieved. The test results show that the above-mentioned method has an obvious effect on improving the classification.(4) Based on the needs of sub-dividing the chamber angle, this thesis puts forward a multi-level assessment method for glaucoma images. To describe the Region of Interest from gradient, EDGELET feature and COHOG feature can improve the differentiating information of all kinds of images and provide more reliability for multi-classification. In the meanwhile, this thesis adopts binary-tree SVM multi-classification algorithm. Based on this method and using the SVM, images are classified into five levels. The test results show that the effects of using this classification method in high-quality photographs are good. |
PDF Full Text Request