The Research on Image Adaptive Block Compressive Sensing Method Based on Underwater Depth Information

Underwater imaging technology has been confronted with the challenge in computation, storage and transmission. Compressive sensing with advantage in reducing data redundancy is widely used in underwater imaging. However, compressive sensing with fixed sampling rate restricts reconstruction quality of the primary object. To address this issue, this paper innovatively proposes a compound depth-based adaptive block compressed sensing method (CD-ABCS). The compound depth matrix that is correlated with underwater depth information, image saliency and image variance is used to set sampling rate of the image block. According to the compound depth matrix, the original image is divided into multi-level image blocks to conduct sparse sampling. Reconstructed image blocks are stitched into a complete image. To verify propose method, experiments including method comparison, ablation study and parameter optimization are executed. Experimental results show that, the proposed method is certified to have a significant improvement in image quality by comparing with other adaptive block compressive sensing methods. Specifically, when the global sampling rate is 0.5, the peak signal-to-noise ratio (PSNR) is increased by 1dB, and the structural similarity (SSIM) improves by at least 0.015. Proposed method is capable of enhancing image quality at various global sampling rates.