Adaptive Sampling Rate-Based Compressive Video Sensing Scheme

In recent years, compressive sensing (CS) theory has broken through the bottleneck of the traditional signal sampling theory. It allows acquiring signals at a sampling rate which is much lower than the Nyquist rate, and ensures perfect reconstruction of the measured signals. Since CS combines sampling with compressing, the complexity of the encoder has been largely reduced, and this theory has also been successfully introduced to the study of video coding. This kind of CS-based video coding scheme is often known as compressive video sensing (CVS). To improve the reconstruction quality of the video, the research of CVS mainly focuses on two aspects:how to design an effective sampling scheme and how to design a suitable reconstruction algorithm. This dissertation studies how to adaptively allocate the sampling rate according to the complexity of the video content, so that the quality of the reconstructed video can be improved. The main contributions of this dissertation are listed as follows:1.A sparsity-based adaptive sampling rate CVS scheme is proposed. In such a scheme, the sampling rate is allocated according to the sparsity of each image block. Firstly, the sparsity in discrete cosine transform (DCT) domain for each block is estimated. Then, the classification of image block is determined according to both temporal and spatial correlation. Experimental results show that compared with the existing adaptive rate CVS methods, our proposed one can get about 0.5dB Peak Signal to Noise Ratio (PSNR) increment.2. In order to further improve quality of the reconstructed results generated by the existing adaptive sampling rate CVS schemes, A spatial-temporal correlation-based adaptive sampling rate CVS scheme is proposed, where spatial correlation and temporal correlation are jointly considered to improve the reliability of the block classification procedure. Firstly, image blocks are divided into different types according to the pre-defined thresholds. Then the initial classification is corrected according to spatial correlation to make the final classification decision. Experimental results show that compared with the existing algorithms, the proposed one can get about 2dB PSNR increment at the same sampling rate.3. The existing adaptive sampling rate CVS schemes cannot adaptively allocate sampling rate for each image block under a given target sampling rate for the whole frame. Therefore, An adaptive rate block CVS scheme based on inter-frame correlation is proposed. Firstly, a fixed part of sampling rate is allocated to every block in a frame. Secondly, the variation of a block is estimated according to the pre-sampled measurements, and the complexity ratio of this block is also calculated. Afterwards, the adaptive part of sampling rate is allocated according to the complexity ratio of a block, and the final measurements are formed by combining the fixed part and the adaptive part of measurements. Experimental results show that compared with non-adaptive scheme, the proposed method can get about 1dB PSNR increment.