摘要： 为免除抠像过程中的人工参与步骤，提出一种基于显著度的非监督抠像方法。提取输入图像的显著度图，利用块融合与块过滤方法，获得抠像的参考信息，将图像划分为多个层次进行多分辨率抠像，从图像背景中抠取出前景物体。实验结果证明，该抠像方法在计算速度和抠像效果上均优于非监督的光谱抠像方法。

关键词: 非监督, 抠像, 显著度, 块过滤, 块融合

Abstract: To avoid human supervising in the image matting process, this paper proposes an unsupervised image matting approach based on saliency. A saliency map is extracted from an image, and more accurate information is extracted from the map by block merging and filtering methods. The image is divided into multiple levels for multi-resolution matting and then the foreground object is matted out on the basis of the processed saliency information. Experimental results show that the approach is better than the unsupervised spectral matting method in both computing speed and matting effect.

Key words: unsupervised, image matting, saliency, block filtering, block merging

中图分类号: 

• TP391