基于深度的双尺度前背景分离模型

针对前景物体从图像背景中分离时错误率较高的问题，提出一个新的基于深度的双尺度前背景分离模型，利用图像内在特性(提取为拉普拉斯矩阵)及各物体的空间信息(由扫描设备得到的深度图像)，较好地去除了前背景交界处颜色相似性造成的歧义。实验结果证明，该模型在视觉上能大幅度改进典型前背景分离模型的结果。

Two-scale Foreground-background Separation Model Based on Depth

Aiming at the problem of high error rate when extracting foreground objects from background in an image, this paper proposes a new two-scale matting model based on depth, making full use of intrinsic features of an image(extracted as a Laplacian matrix), as well as space information of objects, which greatly reduces the artifacts that arise from ambiguities near the boundaries of foreground objects and background when they have similar colors. Experimental result proves that the model improves the results of classic matting models by human vision.