Saliency-Driven Image Compression for Mobile Video
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image compression techniques, particularly region-of-interest (ROI) based codecs, face limitations in efficiently compressing high-definition video for mobile devices due to limited wireless channel capacity and are tied to specific shapes and codecs, failing to adapt to arbitrary regions of interest and varying network conditions.
Innovation Solution
The method integrates saliency-driven image retargeting into the compression pipeline by non-uniformly downsampling images based on saliency maps, creating residual images, and encoding these components for transmission, allowing for adaptive compression that prioritizes salient areas and accounts for user-defined regions of interest.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If uniform image compression is applied, then processing simplicity is maintained, but image quality in important regions deteriorates
Solution Approach 1:
The patent applies different compression quality levels to different regions of the image based on saliency maps. Important regions (high saliency) are compressed with higher quality, while less important regions are compressed with lower quality. This resolves the contradiction by making compression quality spatially variable rather than uniform, improving overall image quality perception without uniformly increasing complexity.
Solution Approach 2:
The image is divided into multiple regions based on saliency values, with each region assigned different compression parameters. This segmentation allows the system to apply optimized compression strategies to different parts of the image, improving quality where it matters most while managing overall complexity through region-based processing.
2Productivity
If high compression is applied to reduce bandwidth, then transmission efficiency improves, but image quality deteriorates
Solution Approach 1:
Different compression ratios are applied to different regions based on their saliency. High-saliency regions maintain higher quality with less aggressive compression, while low-saliency regions use more aggressive compression. This resolves the contradiction by optimizing the trade-off between transmission efficiency and image quality in a spatially selective manner.
3Manufacturing precision
If ROI-based compression is applied, then important regions maintain quality, but adaptability to arbitrary regions is limited
Solution Approach 1:
The system uses dynamic saliency computation to adaptively identify and prioritize arbitrary regions of interest rather than relying on fixed or pre-defined ROI shapes. The saliency maps are computed dynamically based on image content, allowing the compression system to adapt to any arbitrary region configuration, thus improving both region quality and adaptability.
4Manufacturing precision
If non-uniform downscaling is applied based on saliency maps, then salient regions are preserved, but computational complexity increases
Solution Approach 1:
Saliency maps are computed in advance during the encoding process to identify important regions before compression. This preliminary action allows the system to plan the non-uniform downscaling strategy beforehand, preserving salient regions while managing computational complexity through pre-computed guidance rather than complex real-time decisions.
Data Source
AI summary
Methods for content-aware image compression are disclosed. One method comprises the steps of non-uniformly downscaling an original input image according to a saliency map, creating a residual image, encoding the residual image and downscaled input image, and transmitting the residual image and downscaled input image. The encoded image components are transmitted to a receiver. Downscaling may be performed using an aspect ratio that is automatically calculated from the saliency map. The saliency map may be based on an algorithm specified at an encoder or on regions of interest selected by a plurality of users of receivers that receive the transmitted encoded image components.