Efficient Image Dehazing Using an Encoder-Decoder Network with Residual Learning

The visual quality of images significantly declines in hazy conditions, resulting in reduced visibility and low contrast. Current methods are insufficient in estimating the transmission map, which is necessary for precise dehazing. In this study, we present an effective framework for image dehazing that improves image clarity by utilizing multi-scale feature extraction and an encoder-decoder network with residual connections. To ensure high-quality restoration, our model uses dilated convolutions to capture both local and global details. PSNR, SSIM, and MSE metrics have been used to validate the efficacy of the suggested approach. Additionally, we assessed the visual quality performance on several standard datasets and showed that our method achieves better structural preservation and improved perceptual quality than current state-of-the-art (SOTA) dehazing techniques for outdoor images.