Combined Discrete Wavelet Transform and Noise-Prior Retinex Model for Noisy Low-Light Image Enhancement

Abstract

Retinex-based low-light image enhancement methods have drawn significant interest due to their excellent performance and extensibility. However, most of them exhibit poor robustness when images are corrupted by noise. The main reason is that traditional Retinex models cannot effectively characterize the coupling between noise, reflectance, and illumination in noisy low-light images. To address this limitation, this paper modifies the existing noise-aware Retinex model to better characterize the noise component during noisy image decomposition. To further model the noise distribution, the discrete wavelet transform (DWT) is employed to construct a noise prior in the wavelet domain, which effectively captures noise information in different frequency subbands. Furthermore, contour, texture, and noise gradient correlations are incorporated as prior terms for the illumination, reflectance, and noise components, respectively, leading to a unified optimization model. In addition, an improved illumination stretching function is designed to enhance image contrast while suppressing halo artifacts and noise. Experimental results demonstrate that the proposed algorithm achieves competitive performance compared with representative Retinex-based methods and state-of-the-art deep learning-based low-light enhancement methods in terms of noise robustness and quantitative performance.