Accelerated Research And Application Of Bilateral Filtering

Both spatial similarity and range similarity are taken into account in bilateral filtering so that the edge can be well maintained while noise is eliminated.Thus,it solves the problem of over-smoothness and loss of detail information caused by the traditional low-pass filter.However,the computational complexity of bilateral filtering is directly proportional to the size of the selected filtering window,resulting in its disadvantages of high computing burden and slow running speed under the condition of selecting a large filtering window.If the complexity can be reduced,not only the computational efficiency of bilateral filters can be significantly improved,but also the computational efficiency of a large class of algorithms based on bilateral filters in the fields of computer vision,computer graphics,and machine learning can be improved.Therefore,the research on algorithms of accelerating bilateral filtering is of great theoretical and practical value.However,domestic scholars have rarely studied this field,and there is still a big gap between the quality of domestic research and the most advanced level in the world.Therefore,this paper is dedicated to the study of effective algorithms for accelerating bilateral filtering,and the main contributions are as the follows:1)A method of accelerating bilateral filtering based on factor decomposition and meanvariance optimization.This algorithm overcomes the constraint of the general framework for accelerating the bilateral filtering.Firstly,the kernel function of the range of the bilateral filter is decomposed by factorization,and the one that causes the serious time consuming of the bilateral filter is found as the approximate objective function.Then using the variational method,the mean square error optimization problem for approximating the objective function is reduced to the singular value problem.Finally,better filtering results are obtained by iterative summation than the most advanced acceleration algorithm.2)Designing acceleration algorithm for bilateral filtering by training the neural network.This paper combines Splatting-Blurring-Slicing(SBS)processing,which is the most widely used bilateral filtering acceleration process,with the deep learning method,and puts forward the design of acceleration algorithm for bilateral filtering by training the neural network.That is,transforming the problem of accelerating bilateral filtering into training the neural network.It can not only guarantee the minimum approximation error in training,but also provide a unified perspective for the acceleration algorithm based on SBS processing,which is easy to understand and apply.3)Combining theoretical research with practical application,the system of fast bilateral filtering is designed and implemented.Users only need to input the filter coefficient and the image or video to be filtered,and then can quickly obtain the filtering result that meets the demand.