融合纹理特征与螺旋状搜索聚类的超像素分割算法及应用

Application of combining texture features and spiral search clustering superpixel segmentation algorithm

针对传统SLIC(simple linear iterative clustering)超像素分割算法没有综合考虑图像的纹理信息特征,导致对边缘信息较强和纹理复杂的图像进行超像素分割时,出现边缘检测不灵敏,分割效果不理想的问题。提出了把原图像先经过噪声抑制提取出纹理特征分量,构建以颜色特性、纹理特征和空间位置特征相融合的相似性度量方法。改进后的方法提高了边缘检测的灵敏度,增强了算法在对边缘信息较强和纹理复杂图像进行分割时的鲁棒性。另外,提出利用螺旋线状的搜索方式进行聚类,加速了算法的收敛速度,提高了分割效率。改进后的方法在BSDS500公共数据集上进行了实验,结果显示改进后的方法在边缘召回率、欠分割错误率、可完成的分割精度以及算法运行时间四项指标上优于传统算法。

In view of the fact that the traditional SLIC(simple linear iterative clustering)superpixel segmentation algorithm does not comprehensively consider the texture information characteristics of the image,which leads to the problems of insensitive edge detection and unsatisfactory segmentation effect when superpixel segmentation is performed on images with strong edge information and complex texture.So we propose to extract texture feature components from the original image through noise suppression,and construct a similarity measurement method that combines color features,texture features and spatial location features.The improved method improves the sensitivity of edge detection,and enhances the robustness of the algorithm when segmenting images with strong edge information and complex textures.In addition,a spiral search method is proposed for clustering,which speeds up the convergence speed of the algorithm and improves the efficiency of segmentation.The improved method is tested on the BSDS500 public data set.The results show that the improved method outperforms the traditional algorithm in four indicators:edge recall rate,under-segmentation error rate,complete segmentation accuracy and algorithm running time.