Enhancing Non-Uniformly Lighted Images: Experiments on BSIMAGE

This repository contains the image dataset BSIMAGE and a set of enhanced versions obtained by using 11 different algorithms,  including Retinex inspired approaches, a bilateral filter, multi-level enhancers, segmentation based enhancers, histogram equalizers, deep-learning based techniques.

BSIMAGE contains 70 real-world color pictures acquired by commercial cameras under uneven illuminations caused for instance by specific camera settings highlighting some objects while darkening the rest, by particular positions and orientations of the light sources in the scene, by backlight or by brilliant lights scarcely spreading over the scene. Typically, these pictures contain very dark and very bright regions (see BSIMAGEExamples.jpg for some examples) which make their enhancement particularly challenging. In fact, one has to improve the quality of the dark areas without over-enhancing the bright ones.

The 11 enhanced versions of BSIMAGE are stored as zipped folders, as indicated in the pdf file icnluded in this repository.Please cite the paper indicated in 'Related Works' when using the images contained in this repository.