Image DeHazing Using Deep Learning Techniques
Ravi Raj Choudhary, K K Jisnu, Gaurav Meena
Abstract
The task of image de-hazing has been a challenge in the field of Computer Vision since its inception. The images captured during adverse weather conditions often appear to be of low quality due to the presence of various atmospheric particles, which results in the haze, fog etc. This, in turn, causes trouble in detecting objects in an image. This causes problems for many computer vision problems that rely on the visibility of these images. In this paper we are implementing a deep Generative Adversarial Network for image de-hazing. The original de-hazing approaches use per-pixel loss, which when calculating creates huge differences even when there is the only difference in a single pixel, even if these images are perceptually similar. The perceptual loss function extracts high level features of images using pre-trained models on ImageNet, which removes the problems of per-pixel loss functions.