Study On The Method Of Image Segmentation Based On PCNN Model

Image segmentation is one of the key techniques of image processing, its goal is to separate the image domain into dissimilar regions, which makes the analyzation easier and more meaningful. With the development of the image segmentation theory, a growing number of image segmentation algorithms have been proposed. At present, the common segmentation algorithms can be divided into two categories:traditional segmentation algorithms and segmentation algorithms combined with specific theories. The approach for image segmentation based on pulse coupled neural network (PCNN) model has become a hot topic with its high accuracy and extensive application. PCNN, a single layer, was simulated the synchronous pulse bursts in the cat visual cortex. The model is a novel neural network, meanwhile it not only has good biological characteristics but also is easily combined with other methods, without any training or learning. So PCNN model has been widely used in image segmentation.In this paper, some typical segmentation algorithms and common evaluation criteria are stated, and the traditional pulse coupled neural network model and its simplified model are introduced. This paper mainly studies the image segmentation algorithms based on simplified PCNN, and focus on how to automatically determine the number of loop iteration algorithm and set the model parameters adaptively. For the above two defects, this paper mainly has done the following aspects of work:Firstly, in order to solve the problem about the automatic loop iterations and the optimal segmentation, a new image segmentation method based on the simplified PCNN and uniformity measure is proposed. By calculating the uniformity measure of the corresponding image at each process of iteration, the optimal segmentation result is obtained when the maximum value of the uniformity measure is achieved. Experimental results show that the proposed method can automatically achieve better segmentation results and has a common adaptability.Secondly, in order to avoid setting the model parameters manually, we take advantage of the artificial fish swarm algorithm (AFSA) whose parameters are determined automatically. In this paper, we propose an automatic image segmentation method based on the simplified PCNN. Experimental results show that the proposed algorithm can segment the images automatically and successfully without manually setting the PCNN parameters.