Research On The Optimization Of Color Consistency Processing And Seamline Determination Of Remote Sensing Image

With the development of remote sensing platforms and sensors,big data has become the basic feature of remote sensing images.At present,the efficiency and automation level of digital orthoimage production are difficult to meet the needs of rapid processing and emergency response.The two bottleneck problems in image mosaicking of remote sensing image,color consistency processing and seamline determination,are researched to improve the quality and efficiency of remote sensing image mosaicking and realize the automation,parallelization and intellectualization of the production of digital orthoimage products.The main works and innovations can be summarized as follows:1.The research background and significance of image mosaicking are introduced.The research status and existing problems of color consistency processing in single image,color consistency processing between multiple images,seamline determination and parallel orthoimage processing are analyzed and summarized.2.In order to solve the problem of color inconsistency in single image,a variational method based on Retinex with hybrid constraints is proposed.The problem of image dodging is converted to solution of the variational model.According to the prior information of the image,the variational model is constructed based on the Retinex theory.The anisotropic and isotropic total variation regularization terms are adopted to constrain the reflectance image and the illumination image,respectively.Then,the solution of the variational model is translated into three sub-problems with split Bregman method and calculated using alternate iteration method to accelerate the computation.The experimental results indicate that the proposed method can effectively correct various types of color inconsistency with preserving the textures and details of the remote sensing image.The proposed method is practical and feasible for color consistency processing.3.Aiming at the problem of color inconsistency between multiple images in the region,a color balancing method based on adaptive block Wallis algorithm is proposed.Firstly,the images are adaptively divided into blocks according to the coefficients of variation.Bilinear interpolation is used to determine the transformation parameters of each pixel.And the Wallis transform is adopted to eliminate the color differences between adjacent images.Secondly,Voronoi diagram is generated to determine the adjacent relation of images.Dijkstra algorithm is employed to calculate the shortest path and determine the processing sequence to control the color consistency of the whole region.The experiments illustrate that the good performance of the proposed method in eliminating the color difference and contrast difference between adjacent images while ensuring the overall color consistency in the region.4.A method of elicitation seamline determination based on the improved A* algorithm is proposed.Firstly,the difference image is generated based on gray and gradient information.The mathematical morphology method is used to optimize the difference image.Secondly,the improved A* algorithm and pyramid search strategy are applied to improve the efficiency of seamline determination.Thirdly,a method of seamline network generation is presented based on the improved A* algorithm and Voronoi diagram for image mosaicking of multiple images in a region.Finally,a distance weight method of seamline removel based on the optimized template for the color transition when the seamline is a curve.Experiments verify the validity and effectiveness of the proposed method,which improve the efficiency and quality of seamline determination.5.The proposed methods of color consistency between multiple images and seamline determination are parallelized based on GPU.The computation and parallelism of the proposed adaptive block Wallis method are analyzed.Then the parallel acceleration strategy based on GPU is studied.The repeated and independent calculations such as bilinear interpolation and Wallis transformation are assigned to the multiple threads of GPU to be implemented simultaneously,achieving the goal of acceleration.The reduction of sum is adopted for the calculations the mean and standard deviation,which are of the low coupling degree.The shared memory is used to perform parallel operation in each thread block.In order to increase the acceleration ratio as much as possible,the configuration partition,memory bandwidth and instruction throughput are optimized according to the characteristics of the proposed method.Experimental results show that the highest speedup of GPU parallel algorithm can be more than 60 times compared with the serial color balancing method based on CPU.In addition,the steps of the optimized A* method are adjusted and parallized to make full use of GPU.Experimental results indicate that the highest speedup of GPU parallel algorithm can be more than 40 times compared with CPU based serial method of seamline determination.