A Sampling-Propagation Matting Method Based on Sample Validity and KNN Classification Labeling

The sampling-propagation matting method has the disadvantage of not separately processing the image region based on the validity of the samples. This paper proposes an adaptive sampling-propagation matting method, which uses the samples validity to combining sampling method or propagation method when matting an image. We begin with using sampling matting method to process the unknown area. And then we propose a comprehensive criterion to judge the samples validity of the sampling matting. If the sampling matting results satisfy the criterion, the results will be output as the first kind of initial matrix elements to following propagation matting method; otherwise, we propose a labeling algorithm base on the KNN classifier to process the area, and the results will be output as the second kind of initial matrix elements propagation matting. The marking algorithm is used for improving the propagation ability and for enhancing propagation matting result. In the following steps, we set different weights value to the two kinds of matrix elements, which will be used as propagation matting method to get the final results. We evaluate the experiment results with the qualitative observation and quantitative analysis. The experiments show that the proposed method has a good results in situation of missing the valid samples and comprehensive image structure, and has a very fast computing speed.