{"title":"基于核分解和自适应混合归一化的条件变分水下图像增强","authors":"Haopeng Zhang , Hongli Xu , Hao Liu , Xiaosheng Yu , Xiangyue Zhang , Chengdong Wu","doi":"10.1016/j.neucom.2025.130845","DOIUrl":null,"url":null,"abstract":"<div><div>Enhancing underwater images has gained significant interest due to its wide range of applications in ocean engineering and marine robotics. However, underwater images often suffer from ambiguous degradations, making it difficult to construct a deterministic map between distorted underwater images and reference images. In addition, the distortions and artifacts affect the quality of underwater images in local details and global structures. To this end, we propose a novel network based on <strong>C</strong>onditional <strong>V</strong>ariational Autoencoder (CVAE) for <strong>U</strong>nderwater <strong>I</strong>mage <strong>E</strong>nhancement, named CVUIE. Specifically, to capture inherent uncertainties in underwater scenes and generate robust enhanced output, we propose a novel network structure that combines the CVAE with adversarial learning. Then, we present a Kernel Decomposition Attention (KDA) module to process and enhance features over a broader respective field. Moreover, to balance the complex details and global structures of enhanced images, we design a Probabilistic Adaptive Hybrid Normalization (PAHN) module. Evaluations conducted on multiple benchmark datasets prove that the proposed network performs qualitatively and quantitatively better than existing state-of-the-art methods. Real-world experiments have also demonstrated the promising future application prospects of our method.</div></div>","PeriodicalId":19268,"journal":{"name":"Neurocomputing","volume":"650 ","pages":"Article 130845"},"PeriodicalIF":5.5000,"publicationDate":"2025-07-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Conditional variational underwater image enhancement with kernel decomposition and adaptive hybrid normalization\",\"authors\":\"Haopeng Zhang , Hongli Xu , Hao Liu , Xiaosheng Yu , Xiangyue Zhang , Chengdong Wu\",\"doi\":\"10.1016/j.neucom.2025.130845\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<div><div>Enhancing underwater images has gained significant interest due to its wide range of applications in ocean engineering and marine robotics. However, underwater images often suffer from ambiguous degradations, making it difficult to construct a deterministic map between distorted underwater images and reference images. In addition, the distortions and artifacts affect the quality of underwater images in local details and global structures. To this end, we propose a novel network based on <strong>C</strong>onditional <strong>V</strong>ariational Autoencoder (CVAE) for <strong>U</strong>nderwater <strong>I</strong>mage <strong>E</strong>nhancement, named CVUIE. Specifically, to capture inherent uncertainties in underwater scenes and generate robust enhanced output, we propose a novel network structure that combines the CVAE with adversarial learning. Then, we present a Kernel Decomposition Attention (KDA) module to process and enhance features over a broader respective field. Moreover, to balance the complex details and global structures of enhanced images, we design a Probabilistic Adaptive Hybrid Normalization (PAHN) module. Evaluations conducted on multiple benchmark datasets prove that the proposed network performs qualitatively and quantitatively better than existing state-of-the-art methods. Real-world experiments have also demonstrated the promising future application prospects of our method.</div></div>\",\"PeriodicalId\":19268,\"journal\":{\"name\":\"Neurocomputing\",\"volume\":\"650 \",\"pages\":\"Article 130845\"},\"PeriodicalIF\":5.5000,\"publicationDate\":\"2025-07-03\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Neurocomputing\",\"FirstCategoryId\":\"94\",\"ListUrlMain\":\"https://www.sciencedirect.com/science/article/pii/S0925231225015176\",\"RegionNum\":2,\"RegionCategory\":\"计算机科学\",\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q1\",\"JCRName\":\"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Neurocomputing","FirstCategoryId":"94","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S0925231225015176","RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"COMPUTER SCIENCE, ARTIFICIAL INTELLIGENCE","Score":null,"Total":0}
Conditional variational underwater image enhancement with kernel decomposition and adaptive hybrid normalization
Enhancing underwater images has gained significant interest due to its wide range of applications in ocean engineering and marine robotics. However, underwater images often suffer from ambiguous degradations, making it difficult to construct a deterministic map between distorted underwater images and reference images. In addition, the distortions and artifacts affect the quality of underwater images in local details and global structures. To this end, we propose a novel network based on Conditional Variational Autoencoder (CVAE) for Underwater Image Enhancement, named CVUIE. Specifically, to capture inherent uncertainties in underwater scenes and generate robust enhanced output, we propose a novel network structure that combines the CVAE with adversarial learning. Then, we present a Kernel Decomposition Attention (KDA) module to process and enhance features over a broader respective field. Moreover, to balance the complex details and global structures of enhanced images, we design a Probabilistic Adaptive Hybrid Normalization (PAHN) module. Evaluations conducted on multiple benchmark datasets prove that the proposed network performs qualitatively and quantitatively better than existing state-of-the-art methods. Real-world experiments have also demonstrated the promising future application prospects of our method.
期刊介绍:
Neurocomputing publishes articles describing recent fundamental contributions in the field of neurocomputing. Neurocomputing theory, practice and applications are the essential topics being covered.