.png){kind=icon}
.png){kind=icon}
Return
A matting method based on full feature coverage
.png){kind=icon}
OA
AI
DOI：10.1007/s11042-018-6690-1.png)
.png){kind=icon}
.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
The sampling-based matting method is an important method for image matting. There are three key techniques in sampling-based matting: 1) how to build a sample-set; 2) how to travel a sample-set; 3) how to obtain a good sample-pair. Although sampling range has expanded from local to global, the existing approaches to build the sample-set are still limited within the boundary areas of a trimap. Therefore, some valid samples may be ignored if they are far away from the trimap boundary. The so-called global samplings are limited by this disadvantage. Our idea comes from the observation that the samples on both sides of a image edge of the whole image are most representative. Furthermore, in the color space, the pixels in the smooth region are very close to the pixels near the image edge. Based on the discoveries, we present a full feature coverage sampling method, which utilizes the edges as clues to search all possible samples of the whole image area. First, we adopt edge detection to find the edges of the image. Second, the pixels near the edges are gathered into the sample-set. Third, because the population of a complete sample-set is much larger than existing sample-set, we propose an optimization approach to accelerate travelling sample-sets. Fourth, we propose a selective strategy and adopt a propagation matting to enhance the results of sampling matting. Finally, the experimental results are tested on an online benchmark. The results show that the proposed method outperforms many other sampling-based matting methods. The ranking of our method is at the forefront.
Keywords:
Image matting
Edge sampling
Sample-set reduction
Search strategy
Selection criterion
Propagation matting
AI Summary
Key information extracted from the uploaded paper, including a brief overview, abstract, background, key highlights, visual analysis, and future outlook.
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W