The Research And Realization Of Image Segmentation Based On Artificial Life

Image segmentation is the base of image analysis, image recognition and the image understanding. Image segmentation is a technique which divides an image into some special areas and gets interesting areas. Many researchers have been working on it for a long time. Main difficulties or obstacles to image segmentation are the changing of image and the noise. So far there have been many image segmentation algorithms, however, each of them has its own problems that impact the algorithm's performance and application. So we need to get more new methods and makes an intensive study of it.Artificial Life is a new research area which is transdisciplinary. It shows its potential superiority on solving complex problems. If we apply the artificial life to the image segmentation, there would be more extensive research space and good application prospects. And maybe we will find a lot of novel and much better methods. This paper puts forward two artificial life modes based on frozen picture and video frequency sequence respectively. The first is an artificial life model based on the Cellular Automata, the other is a ALife mode which has a variety of size. In the first ALife model we take the image as the environment of the agents. Through some living action like propagation, death, expansion, moving and so on we finally get the result of image segmentation.In the multi-granular ALife model which is extracted according to video feature of video sequence, we also take the image as the environment of agents. It's not a single image at this time but a serial of images of a video. The difference of the images is the environment changing. Agents can live by getting the changing energy. They can detect the changing nearby and extensive to that direction. And in this model some small agents can aggregate a cluster as a big agent. Both of modes have their rules which every agent must observe. Under the rules each agent can choose its next action. The two models have bottom-up and non-overall control features. Through the results of experiments, we conclude that both models have a good performance and application prospect.