Image Dehazing using Dark Channel Prior

Abstract

Windstorms, foggy winters, and sand storms generally degrade image quality and have an impact on computer vision applications, which could be a safety concern for drivers because of light dispersing and retention by dust remnant or air molecules, making the view hazy or foggy. Our method relies on the Dark Channel Prior (DCP) methodology, which improves visual contrast too. Upon quantifying the ambient light, the value is utilized for calculating transmission estimate. The transmission map is fine-tuned afterwards, and even the image is recovered. By processing locally, we simplify the refining stage and localize the refinement zone by working with the coarse dark channel. This strategy will reduce blurriness while restoring the true brilliance of distant objects. To compare we have chosen different existing methods like All-in-One Dehazing Network (AOD-net), Convolutional Neural Network (CNN), High Resolution de-hazer(HR de-hazer) and make the comparison using metrics like Peak Signal to Noise Ratio(PSNR) and Structural Similarity Index (SSIM). Not only in terms of quality parameters but we have also checked over the latency period where we found DCP is giving a valuable output. Hence, the Peak Signal to Noise Ratio value of our proposed method is 20% greater than that of existing methods.