Advanced Natural Color Image Segmentation Based On Region Growing And Clustering Algorithm

Segmentation plays an important role in image analysis. Image segmentation refers to partitioning an image into different regions that are homogeneous or "similar" in some image characteristic. It is an important first task of any automated image analysis process because all subsequent tasks, such as feature extraction and object recognition, rely heavily on the quality of the segmentation. So image segmentation is very important. This dissertation work focuses on phase congruency and advanced region growing algorithm. Several aspects were probed into as follows.Firstly, based on the edge detection analysis, a novel and powerful edge detection, called phase congruency, is presented in this dissertation. Compared with all the previous edge detection strategies, which aimed at the gray image, the phase congruency can get more accurate edges of colorized images. The experimental results show that phase congruency algorithm is more accurate than the canny algorithm.Secondly, an advanced region growing algorithm for image segmentation, is presented in this dissertation. The algorithm processes the pixel around the seed instead of four pixels around the seed, so it can save much time with the almost same result by region growing. The experimental results show that advanced region growing algorithm is efficient when it get the same result with the conventional region growing algorithm.Finally, we process the colorized image that is gained from the advanced region growing algorithm to use region clustering. The object of the image that is gained from the advanced region growing algorithm may be not the whole object. In this dissertation, the colorized image gained from the advanced region growing algorithm is clustered. The segmented regions are clusteredusing K-means algorithm. Then we can get the whole object. The experimental results the image clustered is more meaning to us.