Improving Color Modeling for Alpha Matting

Abstract

This paper addresses the problem of extracting an alpha matte from a single photograph given a user-defined trimap. A crucial part of this task is the color modeling step where for each pixel the optimal alpha value, together with its confidence, is estimated individually. This forms the data term of the objective function. It comprises of three steps: (i) Collecting a candidate set of potential fore-and background colors; (ii) Selecting high confidence samples from the candidate set; (iii) Estimating a sparsity prior to remove blurry artifacts. We introduce novel ideas for each of these steps and show that our approach considerably improves over state-of-the-art techniques by evaluating it on a large database of 54 images with known high-quality ground truth.

High-Accuracy and Quick Matting Based on Sample-Pair Refinement and Local Optimization

“…Wang [13] collected dense samples on the basis of the minimal Euclidean distance. Opposed to that, Rhemann [14] collected samples according to the minimal geodesic distance [8]. However, adjacent samples have similar color values which are redundant to estimate alpha values.…”

“…Thus, inaccurate estimations are achieved by sample-pair initialization [13], [14]. Gastal [15] selected the most confident sample-pairs from ones in the neighborhood to refine the current sample-pair.…”

“…Wang [13] firstly exploited the color-sampling strategy to estimate alpha values, followed by the random walk optimization [18] to smooth matting results. Opposed to the Euclidean distance [13], Rhemann [14] introduced the geodesic distance to gather samples. Additionally, the alpha matte was formulated as a Markov Random Field (MRF) model and optimized globally.…”

Confidence-driven image co-matting

“…Hybrid methods [2,3,4,5] combine the above two approaches, assuming that sampling errors can be eliminated under smoothness constraints, while affinitybased propagation error can be suppressed by accurate color sampling. The recent work by Chen et al [13] integrates sampling-based priors with both local and nonlocal affinities together to achieve the best performance on the benchmark dataset [14].…”

“…Recent sampling-based methods follow this framework but seek to expand the sampling space in a more accurate and efficient way. Rhemann et al [3] collect nearby samples in geodesic distance instead of Euclidean distance. Shared matting [4] shoots rays in several directions and samples known boundary pixels on the rays.…”