Research On Light Visual Feature Extraction And Matching Method For Underwater 3D Reconstruction

The ocean gradually become the second largest strategic development space for human because of containing rich biological resources,mineral resources,oil resources and so on.Machine vision play an more and more important role in the marine resources exploration work because of high precision and abundant information advantages.However,the application of machine vision in the water has been limited due to the ambiguity and complexity of the underwater environment,such as underwater images with low contrast,and seriously affected by noise,greatly decreased image quality.The above problems make it difficult to post-process the underwater image.This paper presents a preprocessing method for solving the problem of underwater image imaging with binocular stereo vision technology as the research object aiming at the problem of underwater imaging and based on the research of various image processing methods.After depth understanding of image texture features,texture features are extracted from underwater images and used in local stereo matching methods.Finally,the local stereo matching method based on multi-feature fusion is applied to the underwater.The main work of this paper is as follows:(1)The binocular stereo vision system is studied firstly in this paper,the camera model,the basic principle of binocular stereo vision and the camera calibration are introduced.And the principle and classification of stereo matching are introduced in detail.(2)Because of the absorption and scattering of light caused by water and a large number of suspended particles and microorganisms,the propagation distance of light in water becomes shorter,underwater image's contrast was obviously decreased and affected by noise seriously.After the study of a variety of image preprocessing methods,we propose a method combining the guided filter and the contrast limited adaptive histogram equalization(CLAHE)to deal with the underwater image,the experimental results show that the underwater image quality can be improved by using this method.(3)The description method of image texture feature are studied,the texture features of the underwater image are extracted by using the gray level co-occurrence matrix and the fractal dimension.Angular Second Moment,Correlation,contrast and entropy were extracted in the four directions of 0°,45°,90° and 135° by using the gray level co-occurrence matrix.Mean and variance of the feature are calculated and combined with the image fractal dimension to form the texture feature vector.Then the genetic algorithm is used to optimize the obtained feature vector to reduce the redundancy of the eigenvector.(4)The local stereo matching algorithm is studied,and a local stereo matching method based on multi-feature fusion is proposed.Firstly,we calculate the matching cost based on the image color feature,the gradient feature and the texture feature respectively.Then the three matching costs are weighted fusion to get a new matching cost.The guided filter is used in the matching cost aggregation to improve the efficiency of the algorithm.The proposed algorithm is applied to underwater,and it can get a better disparity map.