Tuesday 04 March 2025
Recently, a team of researchers has made significant progress in developing an innovative image compression technology that can efficiently capture and convey both human-perceived visual quality and machine-analyzed information simultaneously.
The new approach, called Unified and Generalized Image Coding for Machine (UG-ICM), is designed to bridge the gap between human perception and machine analysis by introducing a conditional decoding strategy. This allows the compressed bitstream to be decoded into different versions tailored to either human or machine preferences.
Traditionally, image compression has focused on either preserving human-perceived visual quality or optimizing machine-analyzed information. However, with the rapid growth of artificial intelligence (AI) applications, there is an increasing need for a single image compression technology that can effectively balance both aspects.
UG-ICM achieves this balance by incorporating a novel conditional decoding module into its architecture. This module takes human or machine preferences as extra conditions and uses them to selectively allocate bits during the compression process. As a result, the compressed bitstream can be decoded into different versions with varying levels of detail, texture, and semantic information.
The UG-ICM technology was tested on a range of image datasets and evaluated using various metrics for both human-perceived visual quality and machine-analyzed performance. The results showed that the proposed approach significantly outperformed traditional compression methods in terms of its ability to balance both aspects.
For example, in an object detection task, UG-ICM achieved an average gain of 3% in bits per pixel (bpp) compared to a baseline method, while maintaining comparable perceptual quality. Similarly, in a semantic segmentation task, the technology showed an improvement of 6% in mIoU (mean intersection over union) and 4% in bpp.
The implications of UG-ICM are far-reaching, as it has the potential to revolutionize various applications where both human-perceived visual quality and machine-analyzed information are crucial. For instance, in medical imaging, UG-ICM could enable efficient compression of medical images while preserving critical diagnostic information for doctors.
In addition, the technology can be applied to other areas such as autonomous driving, surveillance systems, and social media platforms, where high-quality images with rich semantic information are essential for machine analysis.
Overall, the development of UG-ICM represents a significant breakthrough in image compression research, offering a novel solution that can efficiently balance both human-perceived visual quality and machine-analyzed performance.
Cite this article: “Unified Image Compression Technology Achieves Breakthrough in Balancing Human Perception and Machine Analysis”, The Science Archive, 2025.
Image Compression, Unified And Generalized Image Coding For Machine, Ug-Icm, Human-Perceived Visual Quality, Machine-Analyzed Information, Artificial Intelligence, Object Detection, Semantic Segmentation, Medical Imaging, Autonomous Driving.
