Cascade Image Matting with Deformable Graph Refinement

Abstract

Image matting refers to the estimation of the opacity of foreground objects. It requires correct contours and fine details of foreground objects for the matting results. To better accomplish human image matting tasks, we propose the Cascade Image Matting Network with Deformable Graph Refinement, which can automatically predict precise alpha mattes from single human images without any additional inputs. We adopt a network cascade architecture to perform matting from low-to-high resolution, which corresponds to coarse-to-fine optimization. We also introduce the Deformable Graph Refinement (DGR) module based on graph neural networks (GNNs) to overcome the limitations of convolutional neural networks (CNNs). The DGR module can effectively capture long-range relations and obtain more global and local information to help produce finer alpha mattes. We also reduce the computation complexity of the DGR module by dynamically predicting the neighbors and apply DGR module to higher-resolution features. Experimental results demonstrate the ability of our Cas-DGR to achieve state-of-the-art performance on synthetic datasets and produce good results on real human images.

PP-Matting: High-Accuracy Natural Image Matting

“…Chen et al [6] and Deora et al [11] use a semantic network to predict trimap and then concatenate it with RGB images as the input of the matting network. Liu et al [24], Yu et al [41], and Ke et al [17] get the alpha mattes in a coarse-to-fine manner. Li et al [20,21] predict the trimap parallel to the alpha matte of the transition region and then fuse them to obtain the final alpha matte.…”

“…Recently, trimap-free approaches have emerged [6,11,17,20,21,23,24,32,[40][41][42], which do not require a usersupplied trimap as the auxiliary input. Lin et al [23] and Sengupta et al [32] replace the trimap with a background image that is easier to obtain.…”