Deep Image Matting

By Ning Xu, Brian L. Price, Scott D. Cohen, Thomas S. Huang. 2017

510 citations recorded by the source corpus.

Image matting is a fundamental computer vision problem and has many applications. Previous algorithms have poor performance when an image has similar foreground and background colors or complicated textures. The main reasons are prior methods 1) only use low-level features and 2) lack high-level context. In this paper, we propose a novel deep learning based algorithm that can tackle both these problems. Our deep model has two parts. The first part is a deep convolutional encoder-decoder network that takes an image and the corresponding trimap as inputs and predict the alpha matte of the image. The second part is a small convolutional network that refines the alpha matte predictions of the first network to have more accurate alpha values and sharper edges. In addition, we also create a large-scale image matting dataset including 49300 training images and 1000 testing images. We evaluate our algorithm on the image matting benchmark, our testing set, and a wide variety of real images. Experimental results clearly demonstrate the superiority of our algorithm over previous methods.

Subfields: Computer Vision and Pattern Recognition

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

Citations in the corpus

FocalClick: Towards Practical Interactive Image Segmentation (2022, neutral): “‘Synthetic’ data uses the datasets of [8,26,43].”
Getting to 99% Accuracy in Interactive Segmentation (2020, methodology): “We constructed our synthetic training dataset in a similar way to the Deep Image Matting dataset from Xu et al. [42].”
Convolution-Weight-Distribution Assumption: Rethinking the Criteria of Channel Pruning (2020, background): “10) ResNet18/34/50 Deep image matting(Xu et al., 2017) Schedule150-255 VGG11/16/19 AlphaGAN matting(Lutz et al.”
Real-Time High-Resolution Background Matting (2020, methodology): “It also requires large volumes of images with high-quality alpha mattes to generalize; the publicly available datasets [32, 24] are too limited.”
Scale-Recurrent Network for Deep Image Deblurring (2018, methodology): “The encoder-decoder structure has been proven to be effective in many vision tasks [23, 33, 35, 41].”
MVSNet: Depth Inference for Unstructured Multi-view Stereo (2018, methodology): “Depth Map Refinement While the depth map retrieved from the probability volume is a qualified output, the reconstruction boundaries may suffer from oversmoothing due to the large receptive field involved in the regularization, which is similar to the problems in semantic segmentation [32] and image ”
Deep Interactive Thin Object Selection (2021, methodology): “Like other synthetic dataset construction pipeline [49], we composite these foreground objects on background images taken from other datasets for training.”
Guided Collaborative Training for Pixel-wise Semi-Supervised Learning (2020, background): “Image Matting [40,47] predicts a foreground mask (matte) from an input image and a pre-defined trimap.”
Concealed Object Detection (2021, unknown): “According to previous studies [59], [60], multistage refinement can improve performance.”
Low-Light Image Enhancement via a Deep Hybrid Network (2019, methodology): “Inspired by the effectiveness of encoder-decoder networks in image denoising [45], dehazing [46], [47], inpainting [48], matting [49], and harmonization [50], we construct our content branch based on the residual encoder-decoder architecture with specific designs for content prediction as follows: 1”
Guided Image Inpainting: Replacing an Image Region by Pulling Content From Another Image (2018, methodology): “Inspired by prior work that shows models trained on synthetic data can generalize well to real images [4, 39], we propose a process to create synthetic training data for our problem.”