Highlight Removel Research On Specular Surface Based On Sparse Representation

In recent years,Structured Light Scanning technology,because of its noncontact,fast measurement and high precision,has been widely used in machine vision,industrial automatic detection,biomedicine,three-dimensional animation and other fields.But in practical work,when measure specular objects by the structure light,the information collected will miss because of the highlight,this has a great impact on the subsequent image processing algorithms,such as,image recognition,subject matching,reconstruction,subject track,etc.This paper around the high light reflecting surface coding measuring this theme,aiming at the problem that reflecting surface code light measurement there is specular which is difficult to remove and image details are losing is studied.Firstly,we study and summarize the theory of highlight remove method at home and abroad.Then analysis of the effect of specular reflection on structural light three-dimensional measurement,and design the general planning of the system based on the theory,containing of the system configuration,coding light selection,platform building.Through the preliminary theoretical preparation,as well as the current research hotspot sparse representation theory related knowledge learning and analysis,finally,we choose Spatially Adaptive Iterative Singular-value Thresholding(SAIST)algorithm to complete the high light suppression,The algorithm has the advantages of improving the precision of the similar information,enhancing the detail information of the image,simplifying the space complexity and time complexity in the iteration process,and removing the highlight pixel adaptively.Based on the analysis of SAIST algorithm,This paper also selects ceramic bottle and ceramic disk with strong reflection property to carry out experiments,the experimental results show that SAIST algorithm can recover the effective information lost in the highlight image,and the highlight area is significantly reduced,and in the three-dimensional reconstruction,the statistical highlight pixels respectively were reduced about 54.88 times and 39.67 times.In addition,compared with the traditional high light suppression method equalization correction method and Coherency Sensitive Hashing algorithm about ceramic bottle and ceramic disk,the results show that the SAIST algorithm results in high light pixels were reduced 10.11?18.69 times and 7.56?13.44 times.So,the algorithm has significant advantage no matter in objective quantitative indicator or subjective vision quality compared with more highlight remove method.