The Study Of Image Segmentation Algorithm Based On PCNN And Otsu Method

Image segmentation is a basic operation, and its main role is to extract the useful part of the image, in order to further analysis and image processing, This is to compress the image data, and is the basis of image analysis. In recent years, scholars have proposed various algorithms to promote the further development of image segmentation, especially in accuracy and efficiency has been greatly improved, but there are still some shortcomings need to be improved. Maximum between class variance(Otsu) is the classical method in the threshold method, and the two dimensional Otsu algorithm has improved the anti noise performance and robustness of one dimensional Otsu. But the efficiency of the algorithm is greatly reduced with the increase of the dimension, which is difficult to be used in real time processing. In addition the pulse coupled neural networks(PCNN) is a new type of neural network model, the strong biological background makes in the field of image segmentation has inherent advantages, the segmentation effect is superior to the traditional algorithms, but there are also many parameters and iteration termination condition is difficult to identify such problems. In this paper, several improved algorithms are proposed, and the effectiveness of the algorithm is verified by experiments.(1) In this paper, the spiral optimization algorithm of simulation natural phenomena is used to improve the computational efficiency of the two-dimensional Otsu algorithm. Compared to other optimization algorithms, there is a better segmentation efficiency, while only set two parameters, and the two parameters are usually not necessary to change, therefore, is a suitable automatic fast segmentation algorithm.(2) In this paper, based on the simplified model of PCNN, to find a more suitable iterative discriminant of the model so that the fitness of the model more widely. The new discriminant is based on the Otsu algorithm, adding the image contrast as the discriminant factor, for the low contrast image segmentation has a good discrimination.(3) For the segmented images not pure enough, existing isolated noise and the edge is not clear these cases, proposed a bidirectional PCNN model that can bidirectional pulse propagation at the same time, not only can capture neighborhood ignition, but also to capture neighborhood without ignition, so that the next iteration will be able to remove noise, and can make the edges more clearly. It can well enhance the final segmentation results.