Research On The Key Technology Of Retinal Image Processing Based On Computer Vision

Retinal image processing based on computer vision is a combination of medical technology and computer science and technology.At present,medical imaging and computer vision technology have been widely used in medical field,and become the key of auxiliary diagnosis and treatment,providing objective basis for clinical medicine diagnosis and treatment.Retinal microangiopathy is a direct response to diabetes,hypertension and cardiovascular diseases.In view of this medical situation,this paper uses computer vision related technology to process the retina image,extracts the features according to the physiological structure of the image,and then digitize the extracted features,so as to realize computer aided diagnosis.It is of great significance to study the key technology of retinal image processing based on computer vision for the realization of computer aided diagnosis in the condition of uneven distribution of medical and medical conditions.In this paper,the key technology of retinal image processing based on computer vision technology is studied.Based on the advanced research theory of computer vision,On the basis of learning and using the computer vision development platform Matlab,this paper takes the retina images in the Drive database and the Stare database as the research object,and explores the structure and characteristics of the image.The main research contents are as follows:(1)Preprocessing of retinal images.The commonly used fundus camera in modern medicine is poor quality due to poor imaging condition and limited natural eye.In this paper,the image in database is enhanced as preprocessing,and a new enhancement method based on dual tree complex wavelet and improved morphology is proposed.Based on the research of image features,we construct sequence of structural elements,and introduce scale factor to achieve multi-scale and proportional enhancement.Compared with the other algorithms,the enhanced image is at least 5%compared with other algorithms,which improves the quality of retinal images.(2)Retinal image field extraction and vascular feature protrusion.Usually,the acquired retinal images include effective information area and invalid information area,so it is necessary to remove invalid surrounding areas.In this paper,we extract the field of view from YIQ spatial luminance information and get effective information area through corrosion operation.Then,the multi-scale enhancementfilter is constructed by combining the Gauss function and the Hessian matrix to enhance the characteristics of the blood vessels,and the maximum filter value is obtained by setting the elastic scale factor.The constructed filter can effectively suppress the non linear part and highlight the linetype part.(3)Multiscale segmentation of blood vessels.According to the histogram curve enhanced by filtering,a multiscale segmentation method of fundus blood vessels based on statistical mixture model is proposed.In this paper,a hybrid model is constructed on the basis of statistics to fit the blood vessels.The traditional statistical model needs to change the type and parameter of the model according to the different image,this paper uses 1 Gauss model and 2 exponential model to make a fixed mixed mode.The model parameters are jointly solved by means of K mean clustering and EM algorithm.The algorithm is evaluated by comparison of accuracy,sensitivity and specificity.The results show that the accuracy of the algorithm is 94.62%.It has a good effect on the extraction of blood vessels from the retinal images.