Research On Video Surveillance Image Processing Algorithm Based On Compressed Sensing

With the development of science and technology,the field of video surveillance technology has been extended to all aspects of people's lives.Due to limitations of imaging equipment or hardware,video surveillance images acquired in harsh conditions often contain noise and low resolution,therefore,the software algorithms for processing low-quality images to obtain high-quality images become the focus of research and attention.Compressed sensing theory uses the properties that signals can be sparsely expressed,with a small amount of observed signals,the compressed signal can reconstruct the original signal with high probability,which has broad application prospect in the field of image processing.In this paper,compressed sensing theory is applied to denoising and super-resolution of video surveillance images in order to solve the problem of obtaining low-quality video surveillance images.This paper studies the key technologies of compressed sensing theory,at the stage of sparse representation,the principle of dictionary construction is analyzed,several representative greedy iterative reconstruction algorithms like OMP,ROMP,CoSaMP,StOMP and SAMP are mainly studied in the signal reconstruction stage,the performance comparison is carried out by Matlab simulation experiment,which lays the foundation for the improvement of subsequent algorithms.In this paper,an improved K-SVD denoising algorithm based on compressed sensing is proposed for noisy images.Firstly,the original noisy image is separated by high and low frequencies,and the Orthogonal Matching Pursuit algorithm was replaced with Batch-Orthogonal Matching Pursuit algorithm based on residual ratio threshold for the high-frequency part of the image to realize the sparse representation and reconstruction,then the high and low frequency part of the image is superimposed to complete the final denoising.In this paper,the improved algorithm is applied to the de-noising of the video image,and the image which contains the face is identified.Compared with the recognition result of the original image,the recognition rate of the proposed algorithm is higher after denoising,which preserves the detail of the image better and the denoising time is significantly shortened.In addition,this paper also proposes an improved image super-resolution algorithm for low resolution image that combines compressed sensing theory and image structure self-similarity.The standard Euclidean distance is used as a metric for similar blocks and joined to K-SVD dictionary training and learning.The improved algorithm is applied to the super-resolution reconstructions of traffic video images,and the license plate recognition comparison experiment is carried out.The experimental results show that the proposed improved algorithm achieves a better super-resolution reconstruction effect and the recognition rate is improved.