Biography
Enrollment Date: 2013
Graduation Date：2017
Degree：M.S.
Defense Date：2016.12.14
Advisors：Songping Mai
Department：Institute of Microelectronics，Tsinghua University
Title of Dissertation/Thesis：Study on the correction and super-resolution algorithm of
side-scan sonar image
Abstract:
Side scan sonar is an important means of detecting the ocean. Because of the large distorted distortion and low resolution, the image content can not be intuitively understood. Therefore, how to make the side sweep map more intuitionistic and accurately reflect the physical form information of the object of interest is urgent to be solved and is an important issue.
Based on the working principle of the side scan sonar, a method of correction, enhancement and super-resolution for the side sweep map is proposed based on the parameters such as position, attitude and velocity of the sonar system. The target area of the acoustic figure is close to the actual size, high resolution and easy to observe by human eye. And the area is easy to segment identification and other follow-up work.
The main innovations of the paper are as follows: 1) Based on the principle of side scan sonar and image formation, a correction method of side-scan image based on position, attitude and velocity of side scan sonar is proposed and implemented which laid a solid foundation for the enhancement and super-resolution of the subsequent acoustic figure. This method is also a prerequisite for future target segmentation recognition. 2) Through the in-depth analysis of the echo characteristics of the target object, and an acoustic graph enhancement method based on the echo value and echo variation law is proposed. The results show that the echo model and the enhancement method are correct, making it possible to super-resolution the target area. 3) Based on the results of preprocessing such as the correction and enhancement of the acoustic image, a texture similarity super-resolution algorithm based on foreground and background features is proposed. According to the foreground background segmentation results of the target region, the local texture similarity super-resolution algorithm is used for the foreground, and the super resolution algorithm combined with the local texture and the subsea texture library is used for the background.
In addition, this paper studies the extension of the super-resolution algorithm for natural images. Based on the super resolution algorithm of texture similarity, the problem of edge blur in natural image is mainly studied. This paper presents a new neighborhood superimposed algorithm, which is simple and has low search complexity and is suitable for hardware implementation. Compared with the traditional super-resolution algorithm, this method has the advantages of clear edges and rich texture, and has obvious improvement in the objective parameters.