基于两段多组件图割的非监督彩色图像分割算法

提出基于两段多组件图割的彩色图像分割算法，以解决因标签过多和噪声导致的过分割和图割算法低效等问题.多组件图割算法分割图像时，把标签相同的区域处理为该标签的多个组件，结合两层高斯金字塔形成两段多组件图割，以减少分割错误和标签数量，提高分割的性能.算法首先提取基于多尺度四元数Gabor滤波的texton纹理特征，并自适应融合颜色特征；然后使用两段多组件图割获取图像的优化分割，其中，为了引导图割优化的方向，在平滑项中引入彩色梯度信息；最后去除分割结果中的弱边界，获得最终的分割结果.实验结果表明，相对于比较算法，新算法的分割性能有明显提升.

Unsupervised Color Image Segmentation Using Two-phase Graph Cuts with Multiple Components

A color image segmentation method based on multi-component graph cuts and a two-level Gaussian pyramid is proposed to address the problems of over-segmentation and lower efficiency caused by noise and large number of labels. When using multi-component graph cuts to segment an image, regions with the same label are handled as the multiple components of the corresponding segment. In order to reduce the number of labels and segmentation errors, the proposed method integrates multi-component graph cuts with a two-level Gaussian pyramid, called two-phase multi-component graph cuts, which can improve the performance of segmentation. Improved multi-scale quaternion Gabor filter based texton texture features are first extracted and adaptively fused with color features, then the two-phase multi-component graph cut technique is used to get optimized segmentation, where the gradient of the image is used to compute the smoothness term to guide graph cuts to the optimum. Finally, the segmentation result is obtained after removing some weak edges from the optimized result with the proposed method. Experimental results demonstrate the good performance of the proposed method.