A Multimodal Feature Representation Model for Transfer-Learning-Based Identification of Images

IF 1.2 4区 综合性期刊 Q3 MULTIDISCIPLINARY SCIENCES National Academy Science Letters Pub Date : 2024-03-28 DOI:10.1007/s40009-024-01402-7
Nupoor Yawale, Neeraj Sahu, Nikkoo Khalsa
{"title":"A Multimodal Feature Representation Model for Transfer-Learning-Based Identification of Images","authors":"Nupoor Yawale, Neeraj Sahu, Nikkoo Khalsa","doi":"10.1007/s40009-024-01402-7","DOIUrl":null,"url":null,"abstract":"<p>Digital image classification assists in distinguishing natural and synthetic images to detect computer-generated objects. However, CGI improvements make it difficult to discern synthetic photos from genuine ones. Researchers suggest multiple deep learning strategies to differentiate these photo sets utilizing thorough feature analysis. These models are either complex or do not handle image sub-components, decreasing efficiency in large-scale applications. These models fail categorically. To address these issues, this work proposes a novel high-density bio-inspired feature analysis deep learning model for natural and synthetic image sub-classification. A YoLo model initially recognizes objects in input image sets. Processed separately, a hybrid LSTM/GRU model predicts high-density feature sets, which are processed by Elephant Herding Optimization (EHO) Models to identify high inter-class variance feature sets. A customized 1D CNN model is used to categorize the desired features into natural and synthetic components. These classification results establish whether the input image is natural, synthetic, or both. In real-time scenarios, the proposed model is able to improve standard classification models with 8.7% greater accuracy, 10.9% higher precision, 3.2% higher recall, and 8.4% higher AUC.</p>","PeriodicalId":717,"journal":{"name":"National Academy Science Letters","volume":null,"pages":null},"PeriodicalIF":1.2000,"publicationDate":"2024-03-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"National Academy Science Letters","FirstCategoryId":"4","ListUrlMain":"https://doi.org/10.1007/s40009-024-01402-7","RegionNum":4,"RegionCategory":"综合性期刊","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"MULTIDISCIPLINARY SCIENCES","Score":null,"Total":0}
引用次数: 0

Abstract

Digital image classification assists in distinguishing natural and synthetic images to detect computer-generated objects. However, CGI improvements make it difficult to discern synthetic photos from genuine ones. Researchers suggest multiple deep learning strategies to differentiate these photo sets utilizing thorough feature analysis. These models are either complex or do not handle image sub-components, decreasing efficiency in large-scale applications. These models fail categorically. To address these issues, this work proposes a novel high-density bio-inspired feature analysis deep learning model for natural and synthetic image sub-classification. A YoLo model initially recognizes objects in input image sets. Processed separately, a hybrid LSTM/GRU model predicts high-density feature sets, which are processed by Elephant Herding Optimization (EHO) Models to identify high inter-class variance feature sets. A customized 1D CNN model is used to categorize the desired features into natural and synthetic components. These classification results establish whether the input image is natural, synthetic, or both. In real-time scenarios, the proposed model is able to improve standard classification models with 8.7% greater accuracy, 10.9% higher precision, 3.2% higher recall, and 8.4% higher AUC.

Abstract Image

查看原文
分享 分享
微信好友 朋友圈 QQ好友 复制链接
本刊更多论文
基于迁移学习的图像识别多模态特征表征模型
数字图像分类有助于区分自然图像和合成图像,以检测计算机生成的物体。然而,CGI 的改进使得合成照片与真实照片难以区分。研究人员提出了多种深度学习策略,利用全面的特征分析来区分这些照片集。这些模型要么很复杂,要么无法处理图像子组件,从而降低了大规模应用的效率。这些模型基本上都失败了。为了解决这些问题,本研究提出了一种新颖的高密度生物启发特征分析深度学习模型,用于自然和合成图像的子分类。YoLo 模型可初步识别输入图像集中的物体。经过单独处理后,一个混合 LSTM/GRU 模型预测出高密度特征集,再由大象放牧优化(EHO)模型进行处理,以识别高类间差异特征集。定制的一维 CNN 模型用于将所需特征分为自然和合成特征。这些分类结果可确定输入图像是自然图像、合成图像还是两者兼有。在实时场景中,所提出的模型能够改进标准分类模型,准确率提高了 8.7%,精确度提高了 10.9%,召回率提高了 3.2%,AUC 提高了 8.4%。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 去求助
来源期刊
National Academy Science Letters
National Academy Science Letters 综合性期刊-综合性期刊
CiteScore
2.20
自引率
0.00%
发文量
86
审稿时长
12 months
期刊介绍: The National Academy Science Letters is published by the National Academy of Sciences, India, since 1978. The publication of this unique journal was started with a view to give quick and wide publicity to the innovations in all fields of science
期刊最新文献
On the Modeling of Two Covid-19 Data Sets Using a Generalized Log-Exponential Transformed Distribution Hypoestes phyllostachya Baker: A New Record of Invasive Alien Plant Species from Uttarakhand, India Comparison of Different Signal Peptide Targeting EGFP Translocation Periplasm in Salmonella Bistorta coriacea (Sam.) Yonek. & H.Ohashi (Polygonaceae): An Addition to the Angiospermic Flora of India Bacterial Wilt Caused by Ralstonia solanacearum: A Potential Threat to Brinjal Cultivated in Sikkim, India
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
现在去查看 取消
×
提示
确定
0
微信
客服QQ
Book学术公众号 扫码关注我们
反馈
×
意见反馈
请填写您的意见或建议
请填写您的手机或邮箱
已复制链接
已复制链接
快去分享给好友吧!
我知道了
×
扫码分享
扫码分享
Book学术官方微信
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术
文献互助 智能选刊 最新文献 互助须知 联系我们:info@booksci.cn
Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。
Copyright © 2023 Book学术 All rights reserved.
ghs 京公网安备 11010802042870号 京ICP备2023020795号-1