Key Technical Research On The Processing And Analysis Of Retinal Image

Medical image processing and analysis plays an important role in image processing and analysis and is generally considered as a very difficult problem. Great development has been made on the vision ability and quality of medical images due to introducing the graphic and imaging technologies in the last two decades. Thus, the diagnosis and treatment of disease has been improved rapidly. The construction of computer aided analysis system of fundus images can be used of quantity measurement of the tissues in fundus images and accurate discrimination between the normal and abnormal images. It can also lead to standardize on the morphological investigation of the fundus system. As a result, large advantages have been made on the inspection of the fundus images.The key technical problems of retinal image processing and analysis have been studied in this paper, which involves the extraction of the retinal image, the measurement of the width of retinal blood vessel, the registration and the smooth mosaic of retinal images. Firstly, in order to effectively reduce the influence of the inherent focus noise in retinal images and the noise originated from other factors, the retinal images are filtered repeatedly in different directions by employing the gray morphology operations such as dilatation, erosion, open and close computation. Meanwhile, the contrast of the retinal images has been enhanced. In this way, the precision of the retinal image segmentation can be greatly improved. Secondly,based on the extraction of the retinal image, the measurement of the width of retinal blood vessel is carried out through four steps. The first step is to decide the direction of the blood vessel by taking into account the centerline of the blood vessel. The second step is to construct the matching template. The third step is to present the Gaussian model that can approximately simulate the gray distribution of the blood vessel. The last step is to solve the width of the blood vessel by applying a local optimization algorithm i.e. BFGS algorithm to optimize the model parameters. Thirdly, a method based on the blood vessel resemblance is proposed for the retinal image registration after considering the feature of retinal images, the precision, speed and practical application of the registration algorithm. In this method, the registration objective function is developed by using the maximal overlap degree of the retinal blood vessel information and its parameters is resolved by introducing a global optimization algorithm i.e. Genetic Algorithm (GA). Finally, the mosaic process of retinal images is implemented through a fade-by-fade method, which can eliminate the mosaic gaps.Based on the theoretical analysis and research, automated extraction of blood vessel is achieved and the performance of the extraction is compared with that of the expert manually depiction. In addition, we have finished the automatic measurement of the width of the blood vessel as well as the automatic registration and smooth mosaic between images. Experimental results indicate that the method is rapid, exact and robust, and can satisfy the practical clinical demand. The retinal image processing system has the advantage of modular structure and universal user interface, which can guarantee the system expand well.