Single Image Dehazing using Alternate Pooling Fused Transformer Block with Attention Network

Author:

M Suganthi1,C Akila2

Affiliation:

1. Kongunadu College of Engineering and Technology

2. Anna University Tiruchirappalli: Anna University - Regional Office Tiruchirappalli

Abstract

Abstract

Haze decreases contrast and limit sight in both outdoor and indoor images. Each pixel's deterioration is unique and is influenced by how far the scene point is from the camera. The transmission coefficients, which regulate the scene attenuation and degree of haze in each pixel, express this dependence. Previous techniques used a variety of patch-based priors and transformers to solve the single image dehazing problem. Although various researches had demonstrated the effectiveness of vision Transformers, our image dehazing method has been able to surpass the state-of-the-art image dehazing networks. As a result, we proposed a novel image dehazing network named Alternate Pooling Fused Transformer Network (APF_TRANS_NET) with Locally Grouped Self Attention. Compared to earlier deep learning-based methods, it performs far better. The proposed approach enhances the ability of vision transformer in Dehazing progress with an efficient transformer along with the dual weighted deep channel and spatial attention mechanism. To show the efficiency of our model, we trained it on five different datasets, including i-Haze dataset, O-Haze dataset, SOTS dataset, RESIDE-6K, and RS-Haze. The proposed our immense model outperforms the prior state-of-the-art techniques, with a significant improvement in its performance. .

Publisher

Springer Science and Business Media LLC

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3