Research On Image Matching Method For The Side-scan Sonar Image

Being the largest unexplored area on earth,the underwater world has great attraction to marine scientists around the world.Due to the physical characteristics of the ocean and the limitations of human beings,people cannot explore the ocean in the same way as they used to on the ground by using optical instruments.Hence,side scan sonar systems have become the most convenient tool for underwater explorations,and have been widely utilized in military and civil fields for seafloor surveying.However,side scan sonar can only present the data displaying in the style of stripe images,it cannot present the whole sea floor terrain in one picture.At the same time,due to the complexity of the underwater environment,side scan sonar images are polluted with serious noises which will result in narrower scope of gray levels and weak features.What's more,multi-direction scanning is usually proceeded for wide accurate terrain images,applying on the same areas that reveal different characteristics with great view point variation.The above mentioned problems had brought about obstacles to image feature matching procedures,which is the key technique of side scan sonar mosaicking.By taking data of real side scan sonar as the research object,the theory and method of side scan sonar image matching technology is studied in depth,and a serious of thorough solutions according to the key problems mentioned above are proposed.In order to effectively solve the above mentioned problems,the procedure of sonar data encoding,image preprocessing,real-time mosaicking,image matching based on features,high noise image matching,image matching under large angle rotation are analyzed in detail and specified realization algorithms are designed and implemented via Visual Studio platform.The main contents of this dissertation include the following aspects:First,side scan sonar data are encoded based on the analysis of the original data,and the nonlinear compensation algorithm of gray level correction is also designed to improve the gray degradation caused by the sound attenuation and direction distortion to enhance the images.The attitude is perturbed when sonar is towed by ships,destroying the sonar images.In order to solve this problem,a course angle optimization method and correction model is proposed to increase the image quality.In this section,classical method of image de-noise is introduced,and an improved BEMD denoising method based on Guided Filter is designed.This method can remove the noise while maximizing the details,laying the foundation for further study.Second,a hierarchical image matching method for side scan sonar is proposed.SIFT algorithm is introduced to show the disadvantage in side scan sonar image matching,because the details are rare in sonar images.Thus,Guide Filter is used to hierarchize image into the layer of details and the layer of backbone.After the enhancement of the layer of details,the features of this layer with the layer of backbone are fused to produce the new features for matching the two images.During this procedure,the location-navigation information is utilized as the assistance layer to decrease the mismatches.The experiments show that this method can obtain better results for side scan sonar images.Third,the dissertation studies the features matching under high noise pollution.The details are destroyed in Gaussian scale space by the Gaussian blur when images suffer serious noise.Conventional methods based on Gaussian scale space do not work in this condition,however,nonlinear diffusion method can make blurring locally adaptive to the image data so that the noise will be removed,but details or edges will remain unaffected.Firstly we improve the PM equation to accelerate convergence in order to protect details and then decrease the dimension of the descriptor to improve the efficiency.Experiments show that as noise goes higher in images,the improved algorithm based on nonlinear scale space remain in better performance.Once again,an algorithm based on nonlinear affine invariant feature is proposed in this section.Multi-direction scanning is proceeded in real sonar applications,so images are affected not only by noises but also by large scale rotation transformations.The same area demonstrates different characteristics under large view point variations so that it is hard to match.Therefore,we first study the theory of affine invariant feature.Then,we introduce affine camera model into nonlinear scale space to design nonlinear affine invariant algorithm NAIR.Experiments under different angle variations show that NAIR outperforms other algorithms for noise sonar images.Meanwhile,to make the experiments close to the actual environment,comprehensive factors including noise,rotation,scale and blur are taken into account.The results show that NAIR algorithm is the best choice so far.Finally,a summary of innovation and research work in this dissertation are provided.Also,the problems required to be solved and works need to be done in the future are discussed.