Fig 1.
Flowchart of the proposed system.
Onboard image compression comprises three stages: saliency detection by CNN generating a pixel-wise saliency map, CTU-level adaptive QP adjustment, and HEVC intra-coding.
Fig 2.
Rate-distortion performance of HEVC, JPEG, JPEG2000, WebP, VP9, and AVC codec over Kodak dataset.
Kodak datasets consist of 25 lossless, true color (24 bits per pixel) PNG images of 768 × 512 pixels.
Fig 3.
Two types of building blocks in ResNet50: (a) convolutional block; (b) identity block.
Fig 4.
The architecture of deep convolutional layers in ResNet50.
Fig 5.
Architecture of proposed neural network for salient region detection.
Fig 6.
QP distribution map.
Fig 7.
Examples of saliency maps over the test images: Original images and corresponding saliency map masks.
Table 1.
Quantitative comparison of our proposed modified residual architecture with 9 traditional bottom-up saliency methods.
Text in bold denotes the best results.
Table 2.
Quantitative comparison of our proposed modified residual architecture with 8 state-of-the-art methods for saliency map extraction.
The best and the second-best results are highlighted in bold.
Fig 8.
Performance comparison with eight classic bottom-up saliency methods.
Fig 9.
Performance comparison of the proposed method with the latest state-of-the-art methods.
(left to right: original image, Ground Truth, ResNet, SKNet, ResNeXt, Res2Net, ResNeSt, ShuffleNet_v2, MobileNet_v3, EfficientNet_b0, and GhostNet).
Table 3.
Performance metrics of proposed algorithm versus the standard HEVC codec under different initial QPs.
Fig 10.
RD curves for the proposed algorithm and the standard HEVC codec for 50 images of the planet Mars under different initial QPs: (a)PSNR (b)PSNR-HVS (c)PSNR-HVS-M (d)SSIM (e)MS-SSIM (f)VIFP.
Fig 11.
Subjective quality comparison between the standard HEVC codec and the proposed adaptive quantization algorithm at different QP setting.
Details are magnified by bilinear interpolation for comparison.
Fig 12.
The figure displays qualitative results comparison at three groups of images when QP is set to 41.
For ease of comparison, details are magnified by bilinear interpolation.