Perceptual stretch and multi-feature fusion for enhancing nighttime images

Haoxiang Lu, Tianle Fang, Zhenbing Liu, Weidong Zhang, Rushi Lan

Abstract: Images acquired under suboptimal illumination conditions encounter low brightness and contrast, blurry details, and color distortion. To tackle these issues, we propose an image enhancement framework based on perceptual stretch and multi-feature fusion, named PSMF. Specifically, a multiscale convolution-guided filter is developed to fully explore hierarchical frequency information at different scales and levels. And an adaptive perceptual histogram equalization relying on frequency clipping technique and visual perception is presented for contrast stretching. Meanwhile, the pyramid transformation is employed for detail boosting. Subsequently, we employ a multiweight map constrained fusion strategy to aggregate these different feature maps, and further remove color deviation according to the characteristic of the image by using a linear color correction model. Extensive experiments on public benchmarks illustrate that our PSMF is superior to state-of-the-art comparison methods in generating artifact-free images with high global contrast, vivid color, and clearer details. The experiments further suggest that our method significantly improves the performance in advanced computer vision tasks, such as feature point matching and face detection.

External IDs:dblp:journals/kbs/LuFLZL25