Research On Multiple Description Image Coding Based On KSVD

Multiple description coding(MDC)is a coding technique that has strong error concealment and error recovery capabilities for pictures and videos.It studies the unreliable transmission caused by the heterogeneity of the Internet and the high bit error rate of the wireless communication network.By dividing the source into multiple independent data streams and using different channel transmissions respectively,the interference and destruction of information transmitted is reduced.Dictionary learning is a method for obtaining a sparse model learning signals with a dictionary.A sparse representation of the input signal is achieved by linearly combining the atoms in the dictionary.In this paper,we study the multi-description image coding based on the dictionary learning algorithm,and the main contributions are as follows:(1)A multi-description image coding based on K singular value decomposition(KSVD)is proposed.KSVD is a relatively efficient dictionary learning algorithm that uses the given training data to obtain an over-complete dictionary,and then learns the transmitted image to obtain its sparse representation.Since the image is a complex two-dimensional signal,fixed transforms such as Wavelet Transform(WT)or Discrete Cosine Transform(DCT)are easy to cause loss of detail information.Therefore,the KSVD algorithm,which can adaptively learn the dictionary,is used to transform the image to obtain a sparse transform coefficient,which can better reflect the details of the image.The efficiency of image restoration using the KSVD algorithm is confirmed in the experimental results.(2)A multi-description image encoding based on quantized KSVD(Q-KSVD)is proposed.Q-KSVD quantifies the atoms of the learning dictionary based on KSVD,compresses the dictionary,and reduces the storage space of the dictionary while maintaining efficient recognition accuracy.In view of the fact that the dictionary atoms in KSVD are simply normalized and stored in space in the form of floating-point numbers,they occupy a large space.Therefore,this paper uses the quantized KSVD to perform sparse transformation on the image,and compresses the dictionary in training to reduce waste of storage space.The experimental results verify that the Q-KSVD algorithm can achieve the same image recovery as uncompressed,and greatly reduce the space occupied by the storage dictionary.