No-Reference Metric Optimization-Based Perceptually Invisible Image Enhancement

This article presents a novel algorithm to enhance the contrast and extract hidden information from perceptually invisible images and overenhanced images with dark regions. The proposed novel algorithm extracts the most significant patch of an image, trains high-boost-emphasis filters on this patch, and based on the optimized trained coefficients enhances the whole image. First, we propose a novel algorithm based on the fusion of simple linear iterative clustering (SLIC) and the Canny edge detection to extract the most informative segment of an image, termed the patch. Second, a contrast enhancement-specific optimization algorithm is designed. The algorithm optimizes no-reference image quality assessment (IQA) performance metrics on the extracted patch, to obtain the most adept filter and its coefficients for the best possible enhancement. The optimized filter coefficients then enhance the illumination and reflectance component of the input image. Extensive evaluation of quantitative performance metrics on three publicly available datasets demonstrates the sublime performance of the proposed algorithm. The enhanced image provides superior perception quality and the best mean value of quantitative parameters compared with state-of-the-art methods.