Several New Methods Of Natural Image Matting

The digital matting science is developed from the film industry. The so-callednature image matting is a technology that extracting the object which the userinterested in. The nature image matting technology is widely used in many fields, forexample, the film and television special effects production, multimedia titles,human-computer interaction and object tracking and so on.This paper proposes several methods that combine sampling and propagation, anda new unsupervised matting method based on saliency detection. The work about thispaper are as follows:1. Proposed a matting method based on kernel function model. This methodestablish mathematical model with the idea of Grab-cut matting method, and use graphcut algorithm to solve this model. This paper gives a general graph cut method and aspeedy graph cut method on the model which we proposed. The proposed method canexact the non-transparent foreground object in nature images, it has a better performand highly computing speed than Grab-cut method.2. Proposed a matting method based on iterative boundary search. This methodcan be divided into four steps. The first one is pre-processing. Second, searchingmatting failed pixels. Third, matting the failed pixels once again. The last is theiterative process. This method can be effectively used to extract foreground objectswith narrow transparent areas from natural images. The proposed method has similareffect with close-form solution matting method while the processing time is only halfof it.3. Proposed a new unsupervised matting method based on the saliency detection.This method establish model use the result of saliency detection as the prioriinformation. The method can automatically extract the foreground object from anatural image without any human interaction. This unsupervised matting method hassignificant improve in matting effect and processing speed than the current mostrepresentive method(spectral matting method) in unsupervised matting.