Non-uniform illumination images are limited visibility due to under-exposure, over-exposure, or a combination of them. Enhancement of this kind of images is a very challenging task in image processing. Although there are lots of enhancement methods to improve the visual quality of images, many of these methods produce undesirable results in the aspect of contrast improvement or saturation improvement. In order to improve the visibility of images without over-enhancement or under-enhancement, a variational-based fusion method is proposed for adaptively enhancing the non-uniform illumination images. Firstly, a hue-preserving global contrast adaptive enhancement algorithm obtains the globally enhanced image. Secondly, a hue-preserving local contrast adaptive enhancement method produces the locally enhanced image. Finally, the enhanced result is obtained by a variational-based fusion model with contrast optimization and color correction. The final enhanced result represents a trade-off between global contrast and local contrast, and also maintains the color balance between the globally enhanced image and the locally enhanced image. This method produces the desirable visual quality in terms of contrast improvement and saturation improvement. Experiments are conducted on a dataset including different kinds of non-uniform illumination images. Results demonstrate the proposed method outperforms the compared enhancement algorithms both qualitatively and quantitatively.

The comparison algorithms include Arici et al.'s contrast enhancement method using Histogram Modification Framework (HMF) [1], Banic et al.’s Smart Light Random Memory Sprays Retinex (SLRMSR)[2], Ignatov et al.'s enhancement method with Deep Convolutional Networks (DeepNet) [3], Fu et al.'s Weighted Variational Model for image enhancement (WVM) [4], Ying et al.'s Bio-Inspired Multi-Exposure Fusion framework for image enhancement (BIMEF) [5], and Tian et al's Global-Local Fusion method for contrast enhancement (GLF) [6].

A collection of the most challenging cases for image enhancement can be obtained from [7]. Each image in this dataset has the regions correctly exposed and other regions severely under-exposed or over-exposed. A good enhancement algorithm should enhance the under-exposed regions and the over-exposed regions. Meanwhile, the well-exposed regions should not be affected.

Executable Matlab codes are available. (Download)

Some examples are shown to subjectively compare the performances of these enhancement algorithms. The contrast improvement and detailed information preservation are considered in the comparisons.

[1]T. Arici, S. Dikbas, Y. Altunbasak, A histogram modification framework and its application for image contrast enhancement, IEEE Transactions on Image Processing 18 (9) (2009) 1921–1935.

[2]N. Banic, S. Loncaric, Smart light random memory sprays retinex: a fast retinex implementation for high-quality brightness adjustment and color correction, JOSA A 32 (11) (2015) 2136–2147.

[3]A. Ignatov, N. Kobyshev, R. Timofte, K. Vanhoey, L. Van Gool, Dslr-quality photos on mobile devices with deep convolutional networks, in: IEEE International Conference on Computer Vision (ICCV), 2017, pp. 3277–3285.

[4]X. Fu, D. Zeng, Y. Huang, X.-P. Zhang, X. Ding, A weighted variational model for simultaneous reflectance and illumination estimation, in: IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 2782–2790.

[5]Z. Ying, G. Li, W. Gao, A bio-inspired multi-exposure fusion framework for lowlight image enhancement, arXiv preprint arXiv:1711.00591.

[6]Q.-C. Tian, L. D. Cohen, Global and local contrast adaptive enhancement for non-uniform illumination color images, in: IEEE International Conference on Computer Vision Workshops (ICCV Color and Photometry in Computer Vision Workshop), IEEE, 2017, pp. 3023–3030.

[7]V. Vonikakis, A collection of the most challenging cases for image enhancement, Accessed on 02-January-2018, https://sites.google.com/site/vonikakis/datasets/challenging-dataset-for-enhancement .