Empirical wavelet transform-based fog removal via dark channel prior

Abstract

Haze and fog removing from videos and images has got massive concentration in the field of video and image processing because videos and images are severely affected by fog in tracking and surveillance system, object detection. Different defogging techniques proposed so far are based on polarisation, colour-line model, anisotropic diffusion, dark channel prior (DCP) etc. However, these methods are unable to produce output image with desirable quality in the presence of dense fog and sky region. In this study, the authors have proposed a novel fog removal technique where DCP is applied on the low-frequency component of empirical wavelet transformation coefficients of the foggy input image. They apply unsharp masking on wavelet coefficients of the embedded wavelet transformed image for improving the sharpness of the output image. Later contrast limited adaptive histogram equalisation technique is used as a post-processing task to the inverse transformed image for producing the sharp and high contrast output. Finally, the colour and intensity of the contrast-enhanced image are uplifted through S-channel and V-channel gain adjustment. The proposed method provides significant improvement to the overall quality of the output image compared to contemporary techniques. The quantitative and qualitative measurements confirm the claims.