Research On Automatic Mapping Technology

Image matting is an image processing technique that extracts foreground objects in any image or video accurately.The technology can be traced back to the early 1850 s optical matting,and now with the rapid development of the computer industry,digital matting replaces the traditional optical matting and become an important field of computer vision.Due to the lack of constraints in the digital matting model,the existing matting algorithm often requires the user to mark the foreground and background portion of the image in advance.And these artificial mark not only increase the complexity of the operation at the same time but also affect the accuracy of matting.In view of the above problems,this paper presents a new method to realize automatic matting based on one single image.This article has done the following work:First,we describe the common matting method in detail.In this paper,the existing matting algorithm is divided into two categories,interactive matting algorithm and automatic digital matting algorithm.In the interactive matting algorithm,we introduce the algorithms of Bayesian matting,Closed-form matting,Graph Cuts,Easy matting and analyze the Learning based matting method in detail.The automatic digital matting algorithm introduce the Spectral matting algorithm,Flash matting,Automatic trimap generation algorithm and Automatic portrait matting,and explain their advantages and disadvantages respectively.Second,we introduce the algorithm of HS method and LK method in detail,and introduce the LK-HS method which is a combination of HS method and LK method and its improved algorithm in detail.Based on the shortcomings of the traditional matting method,combined with the existing optical principle and optical flow mechanism,this paper presents a single image automatic matting technology.We only need a flat mirror and a camera to complete the experimental environment construction.We first take a picture under the experimental environment,and through optical analysis to get one depth calculation equation.And then use the improved LK-HS method to calculate the real and virtual parts' s optical flow of the image,combined with the depth calculation equation to obtain the depth map of target scene,and then the depth map to morphological operation,Finally,it will be calculated based on the learning matting algorithm to generate the final mask,in order to complete automatic matting.