Image Inpainting Method Based Onsparse Bayesian Learning

Sparse representation theory has wide application in image reconstruction,the core of the sparse representation theory is to design a complete dictionary,then reconstruct the training signal through the complete dictionary.however,the traditional sparse representation theory could not fully use the structural information between data,and the sparse bayesian learning method is based on sparse representation theory,which is fully analyzed the structure of the prior information between signal,then to build a complete dictionary to resume training signal.Because of its priori information advantage in data mining,sparse bayesian learning methods become the research hotspot in recent years.Based on the research of bayesian theory,In order to increase the convergence rate of existing beta process factor analysis(BPFA)algorithm,an enhanced BPFA learning algorithm is researched,the K-SVD algorithm is combined with BPFA algorithm when updating the dictionary,which means that when updating parameters,OMP algorithm is used to update dictionary candidate set to increase the convergence rate because K-SVD algorithm is simple and has fast convergence rate.The improved algorithm in this paper can restore different distress image with noise.To test the repair ability and stability of the improved algorithm,experiment selected different distress images to emulate,which add to the different levels and types of noise.Then,analysis algorithm performance by time,PSNR and other data.The results show that the enhanced algorithm could repair the images better and obtain a preferable visual effect.Finally,an enhanced morphological component analysis(MCA)algorithm is researched(MCA-BPFA),the MCA algorithm is combined with the enhanced BPFA algorithm when inpainting.the image is decomposed into texture and structure part using MCA method,then repair different parts by the improved BPFA algorithm.To test the repair ability and stability of the improved algorithm,experiment selected different distress images to emulate.The results show that the MCA-BPFA algorithm could repair the images better.