融合深度信息的Grabcut自动图像分割

Automatic Image Segmentation Combined Grabcut and Depth Information

Grabcut是一种准确度很高的图像分割方法,但是当图像中存在前背景颜色相近、阴影或低对比度区域时,仅利用颜色、纹理等信息难以准确分割感兴趣区域,而深度信息中包含了这些信息所没有的物体相对前后位置的信息.鉴于此,本文在用显著性实现Grabcut自动分割的基础上,融合了深度信息,提高了算法的分割准确度.为了充分利用深度和显著信息,依次从两方面进行了改进:以深度信息指导的显著图来提取Grabcut矩形框;将深度和显著信息通过自适应权重结合到Grabcut的颜色模型中,改进原算法的能量公式.实验表明,与当前主流算法对比,本文算法更有效地结合了深度信息,提高了分割算法的准确性.

Grabcut is a highly accurate segmentation method. However, it is difficult to accurately segment the region of interest using information such as color and texture when there are similar colors between foreground and background, shades or low contrast regions in the image, while the depth map includes the information about the relative front-to-back position of objects. In view of this, Based on the Grabcut automatic segmentation framework, this paper combines the depth information and improves the segmentation accuracy. In order to make full use of the depth and the saliency information, this paper improves it from two aspects： extract the Grabcut rectangle with the saliency map guided of depth information; combine the depth and saliency information into Grabcut＇s color model through adaptive weight. Experiments show that compared with the current algorithms, the proposed algorithm combines the depth in- formation more effectively and improves the accuracy of the segmentation algorithm.