PROJECT TITLE :

Multi-Scale Deep Residual Learning-Based Single Image Haze Removal via Image Decomposition

ABSTRACT:

These images/videos are generally deteriorated by turbid media such as haze, smoke, fog, rain, and snow. Due of the weather, haze is the most typical type of haze in outdoor settings. To remove haze from a single image, a new deep learning architecture (called MSRL-DehazeNet) based on multi-scale residual learning (MSRL) and image decomposition has been proposed in this research. Since the goal is restoration of an image base component rather than end-to-end mapping, most existing learning-based techniques tackle the problem differently. Both our multi-scale deep residual learning and our simplified U-Net learning may be used to remove (or dehaze) the haze from the basic components of a picture, while the detail component can be further enhanced by another taught convolutional neural network (CNN). The feature maps (created by extracting structural and statistical data) and each previous layer may be fully retained and fed into the next layer thanks to the core building block of our deep residual CNN architecture and our simplified U-Net structure. Consequently, there would be no risk of colour distortion in the restored image. Because of this, the dehazed image is created by merging the original hazy image with the improved detail image. Comparing the outcomes of the experiments with those of other methodologies, the proposed framework has been found to be highly effective.