Study Of Image Segmentation Based On Pulse Coupled Neural Networks

With the growing of social needs and the development of science and technology, people pay more attention to scientific research and practical applications in the field of digital image processing. As an important part of digital image processing technology, research on image segmentation has been rapidly developing. But at the same time, in face of the rich image data, different application scenarios, increasing technical requirements and so on, there are more challenges and difficulties in image segmentation field. Pulse coupled neural networks, as a bionic mammalian visual nerve system, is suitable for application in image segmentation. In this background, aim to improve the anti-noise performance of image segmentation algorithm, the accuracy and efficiency of best image segmentation algorithm, this thesis studied on medical fundus image segmentation methods based on pulse coupled neural network. The research contents and innovations are as follows:(1) Considering the weakness of traditional 2D Otsu method in resisting noise, an improved 2D Otsu algorithm based on median filter has been proposed. The effectiveness of this improved method has been demonstrated by large amount of experiments. This method can improve the anti-noise performance and keep more edge details.(2) Incorporating the pixel gray value, neighborhood space information and activate neurons energy feedback, an image segmentation method based on PCNN and the improved 2D Otsu algorithm has been proposed in this thesis. According to the characteristics of vascular network structure of the fundus images, this method segmented the blood vessels network of the enhancement image by using the dynamic ignition characteristics of PCNN, and then used the improved 2D Otsu algorithm to determine the best number of iterations and the best segmentation result. The results, applied to the fundus image, showed that the method has the feasibility and effectiveness.(3) Considering that PCNN has too many parameters and the parameters can’t be automatically determined, a method which using image segmentation quality evaluation method has been proposed. According to PCNN based on the improved 2D Otsu method and the fundus image, using the receiver operating characteristic curve (ROC curve) to explore the relationship between the image segmentation result and parameter setting. This method can improve the scientific ity and accuracy of parameter settings.