Based Compression Video Coding Perception

In recent years, an entirely new signal processing theory, Compressed Sensing orCompressive Sampling, came into being. It can sample and compress signals simultaneouslyat a sub-Nyquist rate and the signal can be reconstructed with a very high precision from farfewer measurements through an optimization method which broke through the traditionaltheory fundamentally. Based on the in-deep research on the Compressed Sensing theory, theresearch in this paper mainly focused on the application of CS theory in2D/3D image coding,the main works and innovations are as follows:1. A novel measurement matrix design and progressive image coding scheme based on theprior information is proposed. As we all know, the traditional measurement matrix we use arerandom matrix which do not consider the characteristic of the signal. In this scheme,considering the distribution characteristic of the image in the transform domain, an adaptiveand stage-wise measurement method is proposed. In the first stage, an original image issampled by the block-based compressed sensing (BCS) method with a fixed measurementmatrix at a low measurement rate, note that only the low frequency components is sampled,then we can get a reconstructed image in the decoder side which be regarded as the base layer.In the decoder side, the DCT coefficients analysis is done for the reconstructed image andthose larger than a threshold are transmitted to the encoder side through a feedback channel asthe prior information to direct the measurement matrix design in the next stage. Then thesecond stage begins, the measurements obtained in the second stage and in the first stage areused to reconstruct the image, thus to achieve progressive coding, and so forth. Experimentalresults show that the proposed scheme achieves a great improvement than the traditionalnon-adaptive method.2. A depth map coding scheme based on compressed sensing with optimized measurementand quantization. In this scheme, the depth map is measured pairwisely with a variabledensity method, then the measurements are quantized with a scalar deadzone quantizer andentropy coded. In the decoder side, considering the sparisity in the pixel domain, a TVconstraint is added in the conjugate gradient algorithm to guarantee the precise reconstructionof the edges. A tradeoff is found between the measurement rate and the deadzone throughexperiments which could optimize the rate-distortion performance. Experimental results show that the proposed scheme can preserve the edges well and the visual quality of the synthesizedview seems better than the JPEG and JPEG2000scheme.