Research On Image Compression And Video Object Tracking Based On Sparse Representation

In recent years,sparse representation theory has become the theoretical basis of many fields.Scholars have been studying the theory of sparse representation,and it has been successfully applied in the field of image processing and pattern recognition.With the acceleration of informatization,static image compression and video target tracking show great value in theoretical research and engineering applications.In this paper,based on sparse representation theory,image compression and target tracking are studied.Due to the reconstruction of the image will be distorted by using the existing image compression standards in the high compression rate,which will affecting the normal use.An image compression method based on sparse representation is proposed in this paper.K-SVD algorithm is used to train the smooth patch and the detail patch samples to obtain the classification redundant dictionary.According to the relationship between the correlation coefficient and the representation error of the atomic and image signals,an improved OMP algorithm is proposed.The improved OMP algorithm is used to represent the image sparsely by using the classification redundant dictionary,and the coefficient of smooth representation and the greater contribution to the image quality are obtained.The nonzero coefficients and the index values corresponding to these coefficients are quantized to realize the image compression.The experimental results show that compared with the existing standard compression method and single redundant dictionary,the proposed method achieves good compression performance at high compression rate,and effectively improves the image quality at high compression rate.Aiming at the problem that L1 tracking algorithm does not consider the background in the tracking process.An improved method based on L1 algorithm is proposed in this paper.Based on the sparse representation theory,target dictionary and background dictionary are used to track target,which trained by K-SVD algorithm.Judging whether the target is occluded before the target tracking,if there is no occlusion,using L1 algorithm to track;when the target is occluded,the target and background dictionary are used to judge whether the candidate is the target.If the candidate is the target,update the particle state,the target and background dictionary,otherwise only update the background dictionary.Experiments show that the improved tracking algorithm can track the target more accurately and improve the accuracy of the algorithm.