Perceptual Enhancement of Low Light Images Based on Two-Step Noise Suppression

Low-light images are seriously corrupted by noise due to the low signal-to-noise ratio. In low intensity, just-noticeable-difference (JND) is high, and thus the noise is not perceived well by human eyes. However, after contrast enhancement, the noise becomes obvious and severe, because JND decreases as intensity increases. Thus, contrast enhancement without considering human visual perception causes serious noise amplification in low-light images. In this paper, we propose perceptual enhancement of low-light images based on two-step noise suppression. We adopt two-step noise suppression based on noise characteristics corresponding to human visual perception. First, we perform noise aware contrast enhancement using a noise-level function. However, the increase of the intensity caused by contrast enhancement reduces JND in low intensity, which makes noise much more visible by human eyes. Second, we perceptually reduce noise in images while preserving details using a JND model, which represents noise visibility in contrast enhancement. We estimate the noise visibility based on the intensity change using luminance adaptation. Also, we extract image details by contrast masking and visual regularity, because textural regions have higher visibility thresholds than the smooth ones. Based on the human visual characteristics, we perform perceptual noise suppression using the JND model. Experimental results show that the proposed method perceptually enhances contrast in low-light images while successfully minimizing distortions and preserving details.