基于模糊估计融合显著性检测的自动抠图算法

提出了一种新的自动抠图算法框架。首先,估计输入图像各个区域的模糊程度;其次,对图像进行显著性的计算;然后融合模糊度和显著性信息,大致分割出前景和背景,从而自动生成标注前景、背景、未知区域的三色图;最后,采用基于采样的抠图算法就可以准确地完成前景目标物体的自动抠图。该算法无须人工辅助或附加信息。在标准数据集和实拍图像上的实验结果证明了该算法的准确性和实用性,可广泛应用于图像和视频的编辑合成。

Automatic matting based on fusion of blur estimation and saliency detection

This paper proposed a new framework for automatic matting of natural images. First, estimated the blurring level of each part of the input image according to the different focusing status in the foreground and in the background. Second, it calculated the saliency for the image and detected the salient regions. Then, with the combined information of blur and saliency, the foreground region and background region could be roughly identified and separated, and the Trimap which denoted the division of the foreground, background and unknown region could be generated automatically. Last, accurate object matting could be accomplished by a sample-based matting algorithm on the basis of the Trimap. This algorithm did not require any user assistance or other additional information. Experiments both on standard data sets and actual captured images well demonstrate the accuracy and effectiveness of this algorithm, which can be widely used in image/video editing and composition.