A Color Image Segmentation Algorithm Based On Seed Region Growing

Color image processing technology has been boomed in recent years, more and more disciplines such as medicine, metallurgy, agriculture, aerospace, transportation, business, entertainment and other fields show the demand of colorized image processing technology. Colorized image segment is a important part of colorized image processing technology, which divides color image into different region based on color, texture and other features. At the same time, color image segmentation is the basis of the following processing. Therefore, it is very important to research on color image segmentation because of its implementation and scientific values.Image segment method based on region growth gets better performance in gray image segmentation, however, it can not select seeds automatically and it has too more seeds due to regarding pixel as a seed. To solve the above problem, a region growth algorithm based on seed regions is proposed in this paper. The method divides the image into closed regions using watershed algorithm firstly. Then, similarity is defined according to the relative Euclidean distance, and Ostu algorithm is taken to select seed regions automatically. Finally, color image is segmented with region growth. Simulation results show that the proposed algorithm get better performance, and can segment image perfectly.On the basis of some traditional image segmentation methods,such as regional grow, threshold segmentation, edge segmentation, character segmentation, a region growth algorithm based on K-means and weighted local features is presented in this paper, which is used to segment color image.The partial similarity characteristic of pixels are used to build color histogram, then K-means clustering algorithm is used to seek the initial seed point. Finally, we calculate the new seed point neighborhood pixel’s feature to obtain the target region and background region. The experimental results show that the proposed algorithm in this paper is more adaptability, more robust and more effective, and it get better results compared with the classical region growing algorithm..