基于多分辨率的分水岭图像分割算法  

Algorithm of Watershed Image Segmentation Based on Multi-resolution

作　　者：查宇飞[1] 牛江龙[1] 毕笃彦[1] 

机构地区：[1]空军工程大学工程学院信号与信息处理实验室,西安710038

出　　处：《计算机工程》2006年第19期202-204,207,共4页Computer Engineering

摘　　要：针对分水岭分割算法的2个缺陷:耗时较长和过分割问题,该算法在低分辨率图像上进行分水岭分割,提高了分割的速度;由低分辨率图像返回到高分辨率图像时,采用了一种基于边缘信息的合并函数,避免了边缘信息的丢失,保证了分割的准确性。该文设计了一种基于梯度图像的噪声抑制方法,可抑制高斯噪声对梯度图像的影响,有效避免了过分割问题。实验结果证明,该算法兼顾了效率和分割的准确性。This paper proposes a novel watershed segmentation based on multi-resolution image, in order to overcome the drawbacks of traditional watershed segmentation： low calculated efficiency and over-segmentation. Watershed segmentation is completed in low resolution to reduce the burden of computer. A new function based on the edges is proposed to merge regions, which can detect the high frequency information lost in low resolution image. An adaptive threshold is proposed for gradient image to denoise. The arithmetic reduces the effect of Gaussian noise to avoid over-segmentation and reduces the burden of merging. Experiments show that the method balances calculated efficiency and segmentation accuracy.

关 键 词：多分辨率 分水岭 小波变换 区域合并 

分 类 号：TN911.73[电子电信—通信与信息系统]