Research And Optimization On Natural Image Matting Method

Accurate extraction of a foreground object from the image is known as alpha matting.It plays a fundamental role in image and video editing operations.The traditional blue matting technology can obtain robust alpha matte based on the assumption of fixed background,but it has strict requirements on experimental conditions.On the contrary,natural image matting has not any restrictions on image background and focus on extracting fuzzy and translucent object.This paper proposed two improved algorithm to solve the existing problems of matting methods,the details are as follows:Firstly,spectral matting method realizes automatic matting based on graph cuts,but its alpha matte is not robust.Based on spectral matting,this paper proposes a saliency-based unsupervised image matting method.The method consists of three steps: Step one,using the saliency detection algorithm to detect the location of the foreground object,the saliency value of each pixel is used as a priori constraint of the matting;Step two,based on spectral clustering and the assumptions of matting model,matting components are obtained by solving the matting equation using Newton iterative method,and classified based on the salient map lately;Step three,alpha matte is the optimal solution of the energy function by random grouping the foreground and unknown matting components.In this paper,the saliency detection algorithm is improved by using the feedback scheme to get more robust results.Experiments show that the proposed method outperforms the methods based on spectral matting both in speed and alpha matte accuracy and is also comparable with the state-of-the-art methods in robustness.Secondly,Spectral matting method obtains matting components by extending the spectral segmentation algorithm and get the alpha matte based on small amount of foreground/background constraints.But when the local color linear model hypothesis invalid or Laplacian failed to clustering the mixed pixels correctly,the precision of alpha matte decline significantly.This paper uses local sampling to estimate the alpha value of the erroneous pixels to optimize the alpha matte and proposes a robust image matting method based on a small amount of manual interaction.The method consists of three steps: Step one,matting components are obtained by using spectral matting algorithm;Step two,the primary alpha matte is obtained based on a small amount of user input.The trimap is constructed by searching the pixels need to be recalculated,which includes the boundary pixels and fuzzy pixel;Step three,the alpha matte is obtained by using sampling estimation and the smoothing function.Experiments show that proposed method can effectively reduce the workload of manual interaction and get a more robust alpha matte than the original method through local sampling optimization.