Ambience Retaining Enhancement Algorithm for Uneven Illumination Images using Nonlinear Pixel Level Intensity Transfer Function (AREA)

• Published in: Journal of Information Science and Engineering Vol. 38; no. 5; pp. 1021 - 1034
• Main Author: 戴安娜．大衛(DAYANA DAVID)
• Format: Journal Article
• Language: English
• Published: Taipei 社團法人中華民國計算語言學學會 01.09.2022; Institute of Information Science, Academia Sinica
• Subjects: Algorithms; Ambience; Color; Datasets; Empirical analysis; Illumination; Image enhancement; Image filters; Luminance; Pixels; Reflectance; Transfer functions; Visibility; Vision systems; Visual perception; prominent luminance level; guided filter; ambience; nonlinear pixel level intensity transfer function; relative visibility
• ISSN: 1016-2364
• DOI: 10.6688/JISE.202209_38(5).0009

Enhancement of uneven illumination images poses serious challenges in image enhancement. This paper presents a simple but effective ambience retaining enhancement algorithm for uneven illumination images based on Retinex theory. The novelty of the method lies in the use of a Nonlinear Pixel level Intensity Transfer function (NPIT) for enhancing the image. Human Vision System (HVS) enhances the scene considering the global and local context of the region. NPIT imitates this functioning of HVS using the parameters prominent luminance level for global information and relative visibility for local information. The image decomposed into illumination image and reflectance image by a Guided Filter. The NPIT mapped illumination image integrate with reflectance image produces the enhanced image. For better visual perception and color reproduction a color balancing is applied as the post processing stage. The algorithm is tested on two publicly available uneven illumination image dataset and a ColorChecker dataset. The empirical results show that the enhanced image of proposed method are naturally looking, artifact free and ambience retaining. The subjective analysis and objective reveal the superiority of the method over other state-of-art methods.