Multiple Description Coding Based On Lattice Vector Quantization

With the continuous development of digital information and multimedia technology,there has been a surge improvement of digital image and video processing technology,especially in the area of stereoscopic images and three-dimensional video,which has become a heating research topic with its ability to interact with users and strong visual impacts.However,with the increasing demands and numbers of users,the amount of data to be processed is far beyond the ability to process.How to transmit compressed stream effectively and reliably under the limited bandwidth has become the key point of image and video processing.Multiple Description Coding(MDC)is to transmit source coding into multiple streams with equal importance information over different channels.At the decoding end,the decoder can reconstruct acceptable low-quality image or video with one description.If all descriptions can be received,the quality of source decoding will reach the best.MDC effectively solves the problem of the quality degradation caused by the network delay when the traditional source is transmitted in a single channel.In this paper,we provide optimization and improvement to the depth image,2D image,and stereoscopic image coding algorithm by considering the error resilience characteristic of Multiple Description Lattice Vector Quantization(MDLVQ)coding visual saliency theory,the completed work mainly includes:(1)Multiple Description Lattice Vector Quantization(LVQ)coding is applied to depth map coding.Different coding schemes are adopted for different sub-bands coefficients after wavelet transform.The low frequency sub-band coefficients are encoded by MDLVQ coding algorithm with a small quantization step,while the high frequency sub-bands coefficients are encoded by improved Set Partitioned Embedded Block Coding(SPECK).The optimized scheme retains all the larger coefficients after the wavelet transform.The experimental results show that the proposed scheme does not cost much bit of coding while obtaining a better reconstruction quality of the depth map and the synthetic view image,especially in the case of unreliable transmission over the network,the scheme has obvious advantages over the single channel 'decoding quality.(2)We proposed an optimized multiple description coding method based on visual saliency.Different from the traditional multiple description coding methods,an adaptive segmentation threshold can be firstly chosen based on the image contents,which can make a good segmentation between the salient regions and non-salient regions.Furthermore,two kinds of quantization steps can be optimally designed for salient and non-salient regions respectively,that is,fine quantization can be realized for salient regions while coarse quantization for non-salient ones.The experimental results have shown although the overall quality of the reconstructed image is comparable to traditional MDLVQ at the same bit rate,both the reconstructed objective and subjective quality of salient regions have been greatly improved.(3)Considering the development of stereoscopic video,a stereoscopic image multiple description coding scheme based on visual saliency is proposed.In acquiring the salient regions of stereoscopic images,not only brightness,color and texture features are considered,but also the depth map as a reference factor is introduced.At the same time,a new feature map fusion method was design.This model combines brightness,color,texture and depth features,which greatly improves the accuracy of salient region detection.Finally,the results of saliency detection are coded by the optimized MDLVQ coding method.Experimental results show that the proposed method has a better rate distortion performance for stereoscopic image.