基于多尺度平滑的前景提取

Foreground extraction based on multiscale smoothing

摘要:

传统的Graph cuts算法可以有效地提取卡通图像前景，但是对自然场景图像效果差.为了提高前景提取的效果，本文提出了基于多尺度平滑的前景提取模型，联合分割和多尺度特征，从适当的尺度特征中提取前景.运用TV保边平滑模型对图像进行平滑，降低了图像区域的非均匀性，保护了边缘，提高了前景提取的效果.实验结果表明，基于多尺度平滑的前景提取算法降低了非均匀区域对前景提取的影响，其评测分数高于传统的Graph cuts算法.

关键词:前景提取;多尺度平滑;全变分保边平滑;Graph cuts

Abstract:

Traditional Graph cuts algorithm can effectively extract the foreground of cartoon images, but satisfactory results are not achieved for natural scene images. In order to improve the performance of foreground extraction, this paper proposes the foreground extraction model based on multiscale smoothing, which combines segmentation and multiscale feature to extract foreground from appropriate scale features. The total variation edgepreserved smoothing model is used to smooth the image, which preserves the edges and reduces the inhomogeneity of the image, finally, improves the performance of foreground extraction. Experimental results shown that the multiscale smoothing based foreground extraction model decreases the negative effect of inhomogeneous regions on foreground extraction, and the evaluation scores are higher than those of the traditional Graph cuts algorithm.

Key words:Multiscale smoothing; Foreground extraction; Total variation edge-preserved smoothing; Graph cuts