Single Image Dehazing Based on Deep Feature Fusion Network
Runjie Yang, Yongde Guo, Gongwang Zheng, Haonan Yang, Yixuan Zhang · 2023
Image dehazing can make images clearer and more realistic. The dehaze image is more easily recognized and understood by the human eye, and also helps to improve the performance of the computer vision system. Transformer has found significant application in image dehazing because of its remarkable ability to extract global features. However, employing Transformer solely in image dehazing often leads to the loss of image details, while CNN is ineffective in global feature extraction. Therefore, we integrated CNN and Transformer to enhance image dehazing. We introduced a novel technique to address the feature fusion dilemma by utilizing a transmission-guided map to extract haze-based photo characteristics, thereby uniting the features of CNN and Transformer. The combined features contain both image nuances and extensive-range data. Comprehensive tests indicate that our approach outperforms all others on numerous quality assessment metrics.