Block Compressed Sensing Of Image Encoding Based On Saliency Information

The compressive sensing breaks the limitation of traditional Nyquistsampling theorem，which can reconstruct original image signal withfewer sampling points than Nyquist sampling theorem required if thesignals are sparse or compressible. Compressed sensing theory hasexpressed good application prospect in image processing, signalacquisition, wireless communication, computer vision and other fields.The image encoding method based on compressed sensing has alsobecome the research hotspot in the current image encoding.This thesis focus on the exploratory study of the image encodingmethod based on block compressed sensing, and take advantage of thesaliency information of the image to optimize the measurement matrixof every image block in order to effectively improve the measurementefficiency in the encoding and the quality of the reconstruction image.This method can fully sample the detail part of the original image. Themainly works and contributions of this thesis are as follows:(1) Saliency-guided compressed sensing approach to efficient laserrange measurement are investigated thoroughly. Considering that thestructured characteristics with high saliency of the image signals are notconsidered into the Gaussian measurement matrices of traditionalcompressed sensing, the saliency-guided compressed sensing approachis applied to the image signals. It takes advantage of such characteristicsto optimize the traditional Gaussian measurement matrices with thedifferent statistical distributions according to different image blocks. Theexperimental results illustrate that this method can effectively improvethe quality of the reconstruction image both in PSNR and subjectivevisual effect.(2) Combined with the saliency information and thinking out that thehuman eyes are more sensitive to low frequency components than to high frequency components, a novel adaptively image coding viacompressed sensing based on saliency information is proposed. Byanalyzing the saliency information of the image, the proposed arithmeticallocates more number of measurement dimension to the regions that thehuman visual system is more sensitive. In this way, it could not onlyrealize the adaptive sampling in the encoding, but also protect the regionof interesting in the original image. The simulation results demonstratethat the proposed method can enhance the quality of the reconstructionimage both in PSNR and subjective visual effect.