The limitation of a low dynamic range in a digital still camera causes coarse color reproduction in darker regions of the output images, requiring contrast correction. Recently, local correction techniques are preferred to prevent unintended contrast enhancement from global correction methods. A multi-scaled retinex is a representative method, resulting in high quality output images. However, the sizes of the Gaussian filters and weights are determined empirically, regardless of the image, even though suitable sizes and weights corresponding to the respective image will induce a better quality. Accordingly, this paper proposes an adaptive multi-scaled retinex using a Gaussian filter set relative to the input image. First, the weight of the largest Gaussian filter is determined by the local contrast ratio from the intensity distribution of the input image. The other Gaussian filters and corresponding weights are then determined using a visual contrast measure(VCM) and halo measure. The VCM is obtained based on the local standard deviation and locally averaged luminance for several test images, while the halo measure is obtained based on the average of the maximum color differences for patches in the Macbeth color checker. Through an analysis of the VCM and halo measure, the sizes and weights of the Gaussian filters are then determined. In addition, the chroma is compensated to overcome the graying-out phenomenon due to a multi-scaled retinex. In experiments, the proposed method was found to improve the local contrast and saturation naturally.
Tae-Hyoung Lee, Wang-Jun Kyung, In-Su Jang, Yeong-Ho Ha, "Local Contrast Enhancement Based on Adaptive Multi-Scaled Retinex using Intensity Distribution of Input Image" in Proc. IS&T 18th Color and Imaging Conf., 2010, pp 365 - 370, https://doi.org/10.2352/CIC.2010.18.1.art00064