Wavelet-based method for enhancing the visibility of hazy images using Color Attenuation Prior

Abstract

Image processing may be used in the majority of computer vision applications, including those for remote sensing, aerial photography, and vehicle navigation. Images taken outside in bad weather could suffer from degradation. Therefore, it is essential for these applications to enhance the degraded images before processing them. The contrast and saturation that are lost due to scattering must be restored to improve the hazy image. To address this issue, a novel model is built for enhancing a single hazy image efficiently and effectively. Using multilevel wavelet transforms, the image is first decomposed into two frequency domains - the low-frequency domain and the high-frequency do-main. Later, it is dehazed only in the low-frequency domain using the Color Attenuation Prior(CAP). Soft-thresholding operation is carried out in the high-frequency domain to remove any residual noise. Additionally, texture detail in this domain can be improved by using the estimated transmission obtained during the dehazing of the low-frequency component of the image by CAP. Extensive tests were performed on images from the RESIDE and HazeRD datasets to assess the efficiency of our method. Results of tests show that our approach dehazes faster and enriches the hazy image more effectively than other existing dehazing techniques.