The Application Of Several Self-organizing Maps In Color Image Segmentation

Image segmentation is the critical problem on image analysis, it is also one of the oldest and the most troublesome problems. The goal is to find the interesting region from the initial images to gain useful information. The most algorithms on color image segmentation are developed based on the gray segmentation, e.g. thresholding method, edge detection and region method.Some methods work well in gray image segmentation, but when applied to color image segmentation, the results are not as perfect as expected. To deal with these problems, we introduce SOM to color image segmentation, the classification results is improved clearly.The main contents of this paper are given as follows:(1) Make a brief description about similarities and differences between gray and color image, and analyze the shortcoming of traditional methods applied in color image segmentation.(2) Introduce the application of several neural networks in color image segmentation. A self-organizing map(SOM) model is analyzed and discussed about the possibilities of application in color image segmentation. The effectiveness is verified by several experiments.(3) Study the similarities and differences between simulated annealing and K-means algorithm, and they are combined into the SOM to enhance the running efficiency of SOM. Experimental results verify the validity of our new method.(4) Research the application of GHSOM in color image segmentation. The experimental results show that the method is very promising.