Efficient naturalistic approach to image contrast enhancement

摘要

Contrast enhancement is an important step in many image processing and analysis applications. Although the idea of enhancing image contrast is simple, most published mechanisms of letting computers do so automatically tend to be complex or not so intuitive, and the performance is generally not consistent across different image shooting conditions. A new naturalistic approach to this problem that directly mimics what a human artist would do for contrast enhancement is proposed. The goal is to make every non-noise detail easily perceivable by a normal human. Specifically, the gradient/contrast and local background are computed for each pixel and then linearly mapped to an objective point in the gradient-background plane. Then, a cross-bilateral filter is used to smooth the two linear map parameter images. A moving local window is used to obtain the local noise level for each pixel, which is used to determine whether the pixel should be enhanced. The proposed approach is easy to implement and leads to superior results compared with typical state-of-the-art methods, as confirmed by experimental results.