基于多尺度特征的Graph Cut图像分割

Multiscale Feature Based Graph Cut Image Segmentation

传统Graph Cut算法忽略不同尺度的特征信息,为了解决这一问题,将多尺度分解模型与Graph Cut相结合提出基于多尺度特征的分割模型。在多尺度分解部分,设计新的变分正则化函数,在平滑非一致性区域的同时保护边缘,对弱边缘有一定的增强效果。在分割部分,根据相邻尺度平滑图像分割结果的相似度设计自动停止迭代分割条件,在迭代分割过程中自动选取具有适当尺度的平滑图像。实验结果表明所提出的分割模型,其分割性能优于传统算法和DEXTR(CNN)算法。

Traditional Graph Cut algorithm ignores the feature information of different scales.segmentation results are sensitive to image features.In order to solve this problem,a segmentation model based on multiscale features by combining multiscale decomposition model with Graph Cut is proposed.In the multiscale decomposition part,a new variational regularization function is designed to smooth the inhomogeneous regions while preserving or even enhancing the edges.In the segmentation part,according to the similarity of adjacent scale smooth image segmentation results,the automatic stop iterative segmentation condition is designed,and the smooth image with appropriate scale is automatically selected in the iterative segmentation process.Experimental results show that the segmentation performance of the proposed segmentation model is better than the traditional algorithm and DEXTR(CNN)algorithm.