| At present, fundus image processing and analysis plays an important role in image processing and analysis, and is generally considered as a very difficult problem. The primary link of the processing and analysis is that, using modern image processing technology improves image quality and highlights the target area, in order to assist clinical diagnoses. Noise and interference unexpected is often contained in the images obtained. Through the application of nichetargeting image processing technology, noise and interference is removed as much as possible, which has adverse effects for the late diagnosis. After that, regions of interest can be segmented efficiently, which is convenient for clinical diagnosis. At the same time, corresponding pathological criteria is given, which is helpful for doctor, to make full understanding of the emerging and developing of fundus lesions. It also can help doctor make a clear identifying between normal and abnormal sign. And then it can greatly strengthen the accuracy and speed of fundus examination. According to the actual fundus images, the technology of preprocessing and vascular segmentation is researched for fundus images in this paper. The main works are as follows.Firstly, the selection of preprocessing algorithm. After preliminary investigation, firstly consider the algorithms which are simple and easy to implement. They are histogram equalization, homomorphic filtering, and median filtering. And make comparative study among them. But the effect is unsatisfactory, which has the problem of noise over-enhancement. Though survey and simulation experiment study, the super-wavelets technology is finally selected to reduce noise in fundus image in the preprocessing stage. But applying super-wavelets technology in fundus image exhibits the"wrap around"effect. This problem can be solved through fusing the Wiener filtering. Experiment result shows that, the method of ridgelet fused Wiener filtering is the most obvious in multi-scale technique. SNR of the image processed by this method improves the most obviously, about 5.5 times compared with the original image, and denoising effect is most obvious.Secondly, effect evaluation of preprocessing for fundus image. In the stage of the objective evaluation, LMLSD algorithm based on Gaussian wave extraction is chosen in this paper, which is employed to evaluate quality of fundus image. This method can effectively avoid the influences of non-uniform blocks. It is closer with visual feeling, compared with the traditional evaluation method.Thirdly, segmentation study of target area. Distributing of blood vessels and the size of pipe diameter in fundus image is an important criterion of auxiliary diagnosis. Based on fundus image preprocessed, preliminary study of image vascular segmentation is given in this paper. On the one hand, this step is a test for effect of image preprocessing; on the other hand, it is prepared for the follow-up image analysis. Two kinds of image segmentation methods are chosen for effective vascular segmentation and detection of fundus image, which are threshold segmentation technology and edge detection technology. In the segmentation stage, it makes contrastive study between them. Threshold segmentation technology adopted in this paper includes iterative thresholding segmentation, thresholding segmentation based on entropy, and Otsu threshold segmentation. Edge detection technology adopted in this paper includes Sobel operator and Canny operator. The results show that segmentation result of threshold segmentation technology cannot guarantee the integrity of the target. However, the effect of improved Canny operator is more obvious, which is edge detection technology. Experiment result shows that this method can ensure integrity and accuracy of the vascular detection for fundus images. And it is rapid, exact and robust. |
