Image Detail Enhancement via Constant-Time Unsharp Masking

Abstract

Image degradation due to weather phenomena and poor lighting conditions is inevitable in photography and computer vision applications. For example, fine details like distant objects or traffic signs are easily obscured by haze, mist, or dust in the atmosphere. Thus, the proposed algorithm is designed to address this issue by means of unsharp masking technique. Our approach differs from other unsharp masking methods in: i) the ability to control the contribution of sharpness enhancement according to the local variance of the image, and ii) the constant-time algorithmic complexity by virtue of the constant-time O(1) box filter. The use of the image's local statistics allows the proposed method to add more details in the heavily-degraded areas and little or no details in the smooth areas. In addition, the advantage of speed facilitates the integration into real-time processing applications. A comparative study and quantitative evaluation are conducted with a state-of-the-art algorithm to demonstrate the performance and efficiency of the proposed approach.