融合多特征的均值漂移彩色图像分割方法

摘要

Since mean shift segmentation algorithm merely concerns about color and spatial information of images, and is not sufficient to segment images with complex texture. A new color image segmentation method using the combination of color, texture and spatial information was presented in this paper. Image feature with color, spatial and texture including polarity, anisotropy and contrast were exacted from images. Segmentation results were obtained by mean shift filtering and merging regions. The experiment results show that the proposed algorithm illustrates excellent performance to natural images with complex texture.%针对均值漂移图像分割方法中只考虑图像颜色和空间信息,对纹理丰富的图像不能进行有效分割的情况,提出一种新的融合图像颜色、纹理和空间等低层特征信息的图像分割方法.用极性、各向异性和对比度来表示图像的纹理信息,并结合颜色和空间信息形成图像分割特征;然后用均值漂移进行图像滤波;最后,进行区域合并得到分割结果.实验结果表明,该方法对纹理丰富的自然风景图像有较好的分割效果.