基于色彩矩的无监督多分辨率图像分割

Unsupervised multiresolution image segmentation based on color moments

无监督的图像分割被广泛的应用于各种不同的条件下,包括数字相机的图像增强,目标识别,基于内容的图像检索和三维图像分析。本文提出了一种新的多分辨率图像分割方法。与其他基于颜色的全局优化分割方法比较,该方法基于人的视觉系统原理,能够将感兴趣的物体从背景中分割出来,同时图像分割是一个多分辨率的分割过程。首先算法在整幅图像中搜索代表物体的特征块,然后利用特征块的色彩矩特征对所有图像块进行聚类,最后对属于物体类的图像块进行高分辨率的分类,直到块中的每一个像素点被区分为背景或物体。实验结果表明,与传统方法相比,本文算法能够在较短的时间内,取得较好的分割效果。

Unsupervised segmentation of images is highly useful in various applications. This paper describes a novel multiresolution image segmentation algorithm. Compared with other colorbased approaches which use global optimization methods, our algorithm could separate a focused object of interestfrom the background based on the principle of human vision system. At the same time, it does perform a multiresolution process in image segmentation. First, the salient block representing object is searched in global image domain. Then, all image blocks are clustered using the feature of color moments in salient block. At last, the algorithm classifies the image blocks of object class in high resolution. Experiment shows that our algorithm achieves better segmentation results at higher speed than the traditional method.