为了弥补活动轮廓算法对自然图像存在欠、过分割现象,根据图像不同区域对分割贡献各异,设计了一个新的保边平滑函数.该函数对不同区域自适应选择扩散方式;结合水平集建立了改进全变分保边平滑分割模型,同时依分割区域的置信度设计了平滑收敛条件,避免了因过度平滑导致分割曲线消失.相对于传统CV和Li算法,该算法分割测评分数提高了3%以上,且对噪声抑制能力较强.%In order to make up the under-segmentation or over-segmentation phenomenon of traditional active contour algorithm in the segmentation results of natural scene images,this paper designed a new edge-preserved smoothing function based on the different contribution to segmentation of different regions.It adaptively selected the diffusion modes for different regions.It built edge-preserved smoothing segmentation model on improved total variation using level set.Simultaneously,to avoid the vanishing of the segmentation curve which caused by over-smoothing,this paper designed the convergence condition of image smoothing according to the confidence level of segmentation sub-regions.It demonstrates that compared to the traditional CV model and Li model,this algorithm raises at least 3% evaluation scores and has stronger ability in noise suppression.
展开▼