Study On The Analysis Of High-Definition Deep-Sea Images Based On Current Image Processing Technology

Because of deep-sea exploration having broad development prospects and higher ROI, the state has increased investment in scientific and technological research in the deep sea in recent years. Getting access to information mostly relies on image processing technology and high-level visual processing system can reliably provide great convenience for deep-sea research. In this paper, we studied deep-sea image preprocessing and early parts of binocular vision system suitable for deep sea environment. And we got the following results.(1) The paper studied binocular vision system based on the principle of parallax and obtained camera parameters by camera calibration. The system model has been analyzed and explained. Typical camera calibration algorithms were summarized. Ultimately, the Zhang calibration method to extract the parameters for solving internal and external corners by checkerboard map was carried out and the calibration results were optimized.(2) A preprocessing process suitable for deep-sea environment was designed after analyzing the optical and imaging characteristics of water. Firstly, a histogram of the image was analyzed to obtain statistical information of the pixel. Secondly, linear transformation of the image not only enhanced the contrast of the sponge body information, but also removed most seamount background. Finally, faced with the remained large amounts of noise introduced by suspended solids, nonlinear median filtering got a good smoothing result.(3) Deep-sea image feature points were extracted. Detailed theoretical analysis on Harris operator, Susan operator, Moravec operator, Trajkovic operator and CSS operator were carried out. And through simulation experiments, extraction effect of the Harris operator, Susan operator and CSS operator were got and compared. The advantages and disadvantages of the various algorithms were summarized and analyzed.(4) Based on experiments, deep-sea images matching under different perspectives were carried out. The principle of image matching and the needed elements were discussed in this paper. Also, many algorithms suitable for the image matching, such as ABS algorithm, SSDA algorithm, NCC algorithm were introduced. The paper explained theoretical analysis and the mathematical expression of the above algorithm. At the same time, some of the algorithms based on sponge images were simulated and compared. Under the condition of the reference with the above algorithms having bad effects, the SIFT algorithm was introduced in the paper. The SIFT algorithm was proved to match the best by simulation with having parallax and ignoring parallax between different scales and images. The extraction accuracy is high and its feature points are representative. The algorithm is very suitable for processing such deep-sea images.(5) The paper also studied the theory of three-dimensional reconstruction and elaborated the epipolar geometry, Fundamental Matrix, Essential Matrix and its derivation relationship between them. Three-dimensional reconstruction process has been summarized to provide a theoretical basis for the later study.