부산대학교 도서관

Backlit image enhancement is a challenging research problem in image processing and computer vision. The complexity of enhancing backlit images is that they simultaneously contain bright and dark intensity regions. Existing state-of-the-art methods usually have low contrast enhancement in the dark regions, are prone to color distortion in the background or noise amplification, and have high computational complexity. Inspired by its potentiality, this paper applies region segmentation and a transfer learning-based approach to enhance the contrast of the dark regions in backlit images using multi-illumination mappings. In the proposed framework, gamma and log transform functions are applied to achieve a varied set of illumination-map to improve the brightness and contrast of the backlit image. The average results of the validation metrics of the proposed method for contrast measure (CM), entropy (E), CM/E ratio, and SSIM are 8.10, 7.52, 1.07, and 0.58, respectively. The experimental results corroborate that the proposed method demonstrates superior performance to state-of-the-art methods both subjectively and objectively. [ABSTRACT FROM AUTHOR]