Efficient and Accurate DWT-based Image Noise Estimation Using Edge, Skewness, and Statistical Information
Yuan-Kang Lee, Jian–Jiun Ding · 2024
In this paper, an innovative algorithm is designed to estimate the variance of image noise accurately and efficiently. Our approach leverages the Discrete Wavelet Transform (DWT) to compute the noise energy of an image, incorporates edge information, and exploits the statistical relationship between the image noise level and the skewness of the noise energy distribution. Our method demonstrates superior performance compared to state-of-the-art algorithms in accurately estimating image noise level under various conditions, including both the well-exposed condition and the low-light condition. Moreover, the proposed algorithm perfectly aligns with the human visual system (HVS). The noise estimation model we designed effectively solves the difficulty of underestimating noise levels in images characterized by high-level noises. The performance of the proposed algorithm is accessed and analyzed across numerous real-world scene images. This extensive evaluation substantiates the practical applicability and robustness of the proposed algorithm.