基于相位一致与区域生长的自然彩色图像分割

提出一种自然彩色图像中有意义区域的分割方法.利用相位一致模型检测图像边界,根据边界颜色在图像空间及色彩空间的联合分布,对长边界进行K-means聚类形成有意义的种子生长点,合并短边界以克服局部细节的干扰,最后区域生长方法作用于色彩空间得到封闭的分割区域.算法克服了传统区域生长算法缺少区域模型的局限性,同时将其成功应用到彩色图像,并得到与视觉判断相一致的有意义的分割结果.实验结果显示了所提方法对于不同自然彩色图像的有效性与适应性.

Natural Color Image Segmentation Based on Phase Congruency and Region Growing

A new method for meaningful regions segmentation in natural color image is proposed in this study.After edge detecting in term of phase congruency,long edge lines are clustered using K means algorithm based on the joint distribution of their color in image space and in color space to get meaningful seeds,while short ones are merged to eliminate the negative affection causing trivial features in images.Region growing technique is employed in color space to achieve the final segmentation results with the properly selected seeds.The proposed method overcomes the disadvantage of traditional region growing method,and applies it to color images.The meaningful segmentation results are consistent with those executed by people and have shown feasibility and effectiveness to various natural color images.