Single Encoder and Decoder-Based Transformer Fusion with Deep Residual Attention for Restoration of Degraded Images and Clear Visualization in Adverse Weather Conditions.

Autor: Shit, Sahadeb, Roy, Bappadittya, Das, Dibyendu Kumar, Ray, Dip Narayan
Předmět:
Zdroj: Arabian Journal for Science & Engineering (Springer Science & Business Media B.V. ); Mar2024, Vol. 49 Issue 3, p4229-4242, 14p
Abstrakt: Removing adverse weather conditions from images, such as haze, fog, rain, and snowfall, is a significant issue in several scenarios. Many techniques have been described in the literature that only involve removing specific types of adverse weather degradation. A convolutional neural network (CNN)-based all-in-one dehaze network was recently presented to remove all adverse weather conditions. But, this method contains many variables because it employs many encoder blocks for each adverse weather removal operation, and its efficiency still has to be improved. This paper concentrates on creating an effective solution to remove adverse weather from the foggy and rainy real-time images. The proposed research presented a single encoder–decoder-based transformer fusion with a multi-head attention module for real-time image dehazing. Also, the proposed method introduces a separated patches module fusion with a deep residual attention module to improve the different weather degradation problems and minimize the feature loss of degraded pixels in the transformer encoder block. The proposed method is validated and tested on real-time foggy and rainy images. The qualitative and quantitative evaluation demonstrates that the proposed method is more efficient than other methods. [ABSTRACT FROM AUTHOR]
Databáze: Complementary Index
Externí odkaz:
Nepřihlášeným uživatelům se plný text nezobrazuje