DCAI-CISNet: Directional Chroma–Adaptive Intensity with Chromatic–Illumination Separation Network for Low-Light Image Enhancement

Abstract

Insufficient handling of color distortions in current enhancement techniques frequently results in color casts and visual artifacts, while reduced visibility, poor contrast, and in-creased noise deteriorate low-light images, compromising percep-tual quality and downstream processing. The majority of LLIE techniques work in the sRGB domain, where brightness artifacts and distortions are caused by a strong relationship between color and intensity, while HSV-based decoupling partially alleviates these issues but introduces noise in low-intensity regions and hue discontinuities near the red boundary. Modern methods such as HVI employ intensity modulation and polarized chromatic rep-resentations to address these limitations. However, their reliance on global or limited adaptive formulations restricts their ability to handle complex spatial illumination variations. To tackle these challenges, we present a novel color representation, termed Directional Chroma–Adaptive Intensity (DCAI), which effectively decouples chromatic and intensity information by suppressing red artifacts and ensuring chromatic continuity by polarizing the Hue–Saturation (HS) plane. Furthermore, we propose a spatially adaptive intensity modulation approach that leverages a network to reduce noise in low-illumination regions. In addition, we design a Chromatic–Illumination Separation Network (CISNet) to accu-rately enhance photometric quality by jointly modeling chromatic and illumination information. Extensive evaluations conducted on diverse paired and unpaired benchmark datasets demonstrate the superiority of the proposed method over fifteen existing state-of-the-art approaches. The implementation code is available at: https://github.com/mousumibehera821/DCAI-CISNet.git.