Saliency-driven adaptive block partitioning with energy-based compressed sensing for underwater image compression

• Saliency-based adaptive block partitioning technique that adaptively adjusts block sizes to preserve structural details in underwater images. • Energy-based compressed sensing approach, allocating sampling rates adaptively based on block energy to improve reconstruction quality. Underwater imaging (UI) generally produces huge volumes of data, as there are high-resolution cameras used for imaging in challenging underwater environments. Underwater communication systems have limited bandwidth and storage capabilities, which makes compression an essential one. In recent times, adaptive compressed sensing (ACS) has emerged as a widely adopted method for image compression. The attractive feature of this ACS is that it can choose varying numbers of image samples from various image blocks. However, the block size remains the same in ACS. This leads to distortion in smooth, low-information blocks after sampling, due to an insufficient sample availability for reconstruction. This problem suggests the use of adaptive block partitioning (ABP), where the entire image is partitioned through clustering of information based on saliency. Accordingly, row-wise and column-wise scanning is performed to group pixels with similar saliency. The total samples to be selected from the image block are then computed using energy-based adaptive compressed sensing (EACS). This results in fair sample allocation and reconstruction across all adaptive blocks in the image. Experimental results show that ABP-EACS achieves smooth reconstruction of underwater image blocks with PSNR increase of 2–4 dB, SSIM improvements of 0.1–0.2, NCC nearer to 1, and NAE nearer to 0.