Research On Noise Suppression Of Infrared Image Based On Sparse Representation

Infrared thermal imaging technology is widely used in various industries because of its anti-interference ability and all-weather imaging.However,since the infrared signal is interfered by the device of the detector itself and is affected by external factors during the transmission process,various noises are introduced in the process of collecting the infrared image by the infrared thermal imaging system,which limits the application of the infrared image.Therefore,it is of great significance to study the infrared image noise suppression method.In order to solve the problem that the traditional infrared image noise suppression method filters out the infrared image noise,the infrared image detail information is also lost.Based on the in-depth study of sparse representation theory,this thesis proposes two sparse representations of infrared image noise suppression methods.Aiming at the problem of poor self-adaptive ability in traditional sparse representation methods,an improved K-SVD dictionary sparse representation infrared image noise suppression method is proposed.The method first performs block processing on the infrared image,each block is a class,and then performs dictionary learning and sparse decomposition for each class.In the dictionary learning stage,the Discrete Cosine Transform overcomplete dictionary is selected as the initial dictionary,and the infrared image block is subjected to multiple singular value decomposition learning to obtain a new dictionary matrix.The discrete cosine transform dictionary is selected as the initial over-complete dictionary for each sample signal,and the image block is sparsely decomposed on the over-complete dictionary by the orthogonal matching pursuit algorithm,and the residual change rate threshold is the sparse decomposition iteration termination condition.Secondly,iteratively update the dictionary to obtain the dictionary matrix and the sparse coefficient matrix of the image block.Finally,the infrared image is reconstructed by using the updated dictionary matrix and the sparse coefficient matrix.Aiming at the problem that the optimized K-SVD dictionary is not structurally strong,a sparse representation of infrared image noise suppression method for non-local clustering dictionary is proposed.The method firstly uses the improved K-means algorithm to cluster the image blocks,and performs principal component analysis on each type of image signals.The feature vectors which can represent the feature information of the image signals are extracted to form the feature matrix,which is used as the initial feature dictionary.Then,K-SVD dictionary learning is performed for each type of image block,and iterations are repeated multiple times to update each type of image region to obtain an overcomplete dictionary matrix and a sparse coefficient matrix.Finally,the dictionary matrix and the sparse coefficient matrix are combined with the index matrix generated during clustering to perform weighted superposition,and the infrared image after noise suppression is reconstructed.The thesis compares the traditional infrared image noise suppression method and the noise suppression method proposed by the paper through Matlab simulation software.The results show that the sparse representation of the infrared image noise suppression method based on the optimized K-SVD dictionary proposed in this paper can suppress the noise in the infrared image more effectively than the traditional method.The sparse representation of the infrared image noise suppression method based on the non-local clustering dictionary can not only effectively suppress the infrared image noise,but also preserve the image detail information,so that the visual effect of the reconstructed infrared image is significantly improved.