Improved Single Image De-Hazing Via Sky Region Detection, Classification and Illumination Refinement
Uche A. Nnolim · International Journal of Image and Graphics · 2019
This paper presents an automated sky detection technique based on statistical and fuzzy rule-based edge detection for improved hazy image contrast enhancement. This is significant since most conventional de-hazing approaches yield hazy images with over-enhanced sky regions and under-enhanced detail regions due to inability to adaptively determine and enhance such regions. Earlier and current schemes developed to remedy this issue are highly complex, usually require training with vast amount of images and manual tuning of one or several parameters. The proposed method utilizes standard deviation and fuzzy logic-based edge detection combined with thresholding algorithms to generate a homogeneity map identifying sky and non-sky regions. The areas of these regions are subsequently computed and used to obtain a homogeneity ratio. The ratio is then used to trigger a decision-based, switching scheme incorporated into a partial differential equation (PDE) de-hazing algorithm to improve results. Alternatively, a log illumination refinement method is proposed as a less complex alternative combined with the modified PDE algorithm to process hazy images without degrading sky regions, while yielding brighter images. Several image datasets from the literature were used to validate the proposed approaches and yielded mostly consistent and comparable results similar to or better than algorithms from the literature.