Research On Key Issues Of Assistant Diagnosis For Cortical Cataract Via Image Processing

The incidence rate of cataract is the highest and the most effective eye blinding disease in the world.Patients often have blurred vision,glare,monocular diplopia,and seriously affect the quality of life.China is a "serious disaster area" of cataract,but the number of ophthalmologists is far from meeting the needs.With the continuous development of machine learning,computer-aided diagnosis system plays an increasingly important role,which provides a new research direction for improving the medical status of cataract.In this thesis,three modules of cortical cataract image preprocessing,cataract lesion detection and lesion classification are studied theoretically.On this basis,according to the characteristics of cortical cataract image,the multi-layer fusion technology and the theory of color constancy are combined to propose an effective accurate cortical cataract image color restoration algorithm;using dark channel and pixel iteration,an effective accurate cortical cataract image color restoration algorithm is proposed A reflective removal method that significantly removes reflections and improves local contrast.In view of the visual significance and self similarity of cataract,a cortical cataract detection algorithm combining fractal and visual significance was proposed.This thesis analyzes and compares a variety of feature extraction and machine learning algorithms,combines color features,texture features and depth features,and uses SVM learning classification model to propose an accurate cortical cataract classification algorithm.The experimental results show that the algorithm can effectively recover the color and remove the reflection of the cortical cataract image,significantly improve the image quality;can significantly improve the shortcomings of the visual significance algorithm to detect the cataract lesions,greatly improve the detection accuracy of the lesions;can effectively accurately and quickly classify the cortical cataract lesions.The main research contents and contributions of this thesis are as follows:(1)The theory of color constancy is deeply analyzed and studied.A color restoration method based on multi-layer fusion color constancy is proposed for cortical cataract image.Compared with common color constancy methods,the experiment shows that the algorithm in this thesis can effectively restore the color of cortical cataract image.(2)In this thesis,the method of reflective removal is analyzed and studied,focusing on the method of reflective removal based on two-color reflective model,and the method of reflective removal of cortical cataract image based on the intensity ratio of dark channel is proposed.Experimental results show that the algorithm can effectively remove the reflection of cortical cataract image.(3)This thesis discusses and analyzes the visual significance algorithm,mainly including the significance algorithm based on contrast and the significance algorithm based on graph model.Aiming at the cortical cataract lesions,this thesis proposes a method combining fractal and visual significance detection.The experiment shows that this algorithm greatly improves the detection accuracy of disease change.(4)A cortex cataract classification algorithm combining manual features and depth features is proposed.The algorithm uses multi-color spatial color moments,texture features extracted by GLCM and MR8,and depth features extracted by CNN to describe cortex cataract images.SVM is used for classification.Experiments show that the algorithm can achieve high accuracy and the calculation time is acceptable.