文献类型：期刊文献

英文题名：Underwater image enhancement method based on a cross attention mechanism

作者：Xu, Sunhan[1];Wang, Jinhua[1];He, Ning[1];Hu, Xin[1];Sun, Fengxi[1]

第一作者：Xu, Sunhan

通讯作者：Wang, JH[1]

机构：[1]Beijing Union Univ, Smart City Coll, Beijing 100101, Peoples R China

第一机构：北京联合大学继续教育学院

通讯机构：[1]corresponding author), Beijing Union Univ, Smart City Coll, Beijing 100101, Peoples R China.|[1141733]北京联合大学继续教育学院;[11417]北京联合大学;

年份：2024

卷号：30

期号：1

外文期刊名：MULTIMEDIA SYSTEMS

收录：;EI(收录号：20240415420373);Scopus(收录号：2-s2.0-85182716344);WOS:【SCI-EXPANDED(收录号:WOS:001144225200006)】；

基金：This work was supported by National Natural Science Foundation of China (No. 62172045 and No. 62272049), the Academic Research Projects of Beijing Union University (No. ZKZD202301).

语种：英文

外文关键词：Underwater image enhancement methods; U-Net; Cross attention transformer; Dynamic enhancement module; Hybrid loss function

摘要：Underwater image enhancement is a technique that improves the quality of underwater images, which makes them clearer and more realistic. However, because of the complexity of underwater environments, underwater image enhancement faces many challenges, such as the variation in underwater optical properties as well as low contrast, low brightness, and color distortion in underwater images. To extract underwater image features more effectively, this paper proposes an underwater image enhancement algorithm called cross attention-based underwater image enhancement (CAUIE). The algorithm combines cross large kernel attention and dynamic enhancement modules to build a U-Net model. Cross larger attention uses large kernel attention mechanism to capture the local and global information of underwater images alternately, thus enhancing the semantic representation of the images. The dynamic enhancement module, by contrast, dynamically adjusts the enhancement parameters according to different regions of the image to acquire detail information. In addition, this paper introduces a contrast regularization loss to construct a hybrid loss function for guiding the training and optimization of the model. The experimental results show that the proposed algorithm outperforms the comparison algorithm in both subjective visual and objective evaluation criteria. Moreover, the proposed model obtains PSNR and SSIM results of 34.86 dB and 0.996, respectively, increasing the results of the previous model by 7.97 dB and 0.099, which illustrates that the proposed algorithm can solve the color distortion problem and recover the contrast and clarity of underwater images.And CAUIE achieved good results in two no-reference underwater evaluation metrics UIQM and UCIQE.