An Adaptive Low-Light Image Enhancement Using Canonical Correlation Analysis

Abstract

Images captured under low-illumination conditions exhibit numerous unwanted appearances such as color distortion, low brightness, a narrow gray range, low contrast, and considerable noise. An excellent low-light image enhancement framework overcomes all undesirable attributes of low-light images. Here, a new fusion technique is introduced based on the canonical correlation analysis of two proposed adaptive Retinex and beta-hyperbolic secant distribution (BHSD) models. Firstly, an adaptive enhanced reflectance image is obtained by a new weighted multiscale Gaussian function and proposed adaptive reflectance map. For the second enhanced image, logarithmic image processing is modified according to the illuminance's factor, and the exponential function is incorporated for contrast boosting. The error function is used for the modified cumulative distribution function of BHSD to obtain the second enhanced image. The proposed framework is compared with state-of-the-art low-light methods for visual and parametric comparison. The results show that the designed framework produces the best results in terms of high contrast, natural brightness, vivid color combination, perfect structure, and texture information with preserving naturalness. Among all comparable methods, the designed technique produces a good result in terms of rich scene details and color reproduction in dark regions.