Semantic image pre-editing improves compression for machine vision tasks
SCOPE: Semantic COdec Pre-Editing for Object-Preserving Compression
Computer Vision and Pattern Recognition
Summary
Images are usually compressed to save space, but standard methods focus on what looks good to humans, not on how well machines can analyze them afterward. The authors created a method called SCOPE that changes images just before compression to keep important parts for machine detection while simplifying less important areas. This method adapts to different compression types without needing separate training. It helps machines detect objects more accurately even when the images are heavily compressed.
What this means in practice
- •For autonomous vehicle engineers: Improve object detection accuracy in compressed images sent from cameras to onboard processors in vehicles with limited bandwidth.
- •For surveillance system developers: Enhance detection reliability in security cameras using standard compression while reducing required storage and transmission rates.
Authors
Silvia Dani, Leonardo Galteri, Marco Bertini
Abstract
Image compression is increasingly used before machine vision systems, yet standard codecs are mainly optimized for human perception rather than downstream task accuracy. We propose SCOPE (Semantic COdec Pre-Editing), a conditional image pre-editor that modifies images before compression to improve the bitrate and detection accuracy trade-off. The model is conditioned on the codec family and compression severity, allowing a single trained network to adapt to multiple compression settings without retraining or loading codec-specific weights. Instead of uniformly preserving the whole image, the pre-editor learns to protect task-relevant regions while simplifying less important background content to improve compressibility. The resulting rate-accuracy curves show consistent gains across different codec families and detector architectures, together with strong transferability to unseen learned and wavelet-based compression families.