融合注意力机制的多特征艺术图像感知方法研究  

Research on multi-feature art image perception based on the integration of attention mechanism

作　　者：何俊杰 胡建君 HE Jun-jie;HU Jian-jun(School of Art,Shanghai Zhongqiao Vocational and Technical University,Shanghai 200000,China;Shanghai Academy of Fine Arts,Shanghai University,Shanghai 200000,China)

机构地区：[1]上海中桥职业技术大学艺术学院,上海200000 [2]上海大学上海美术学院,上海200000

出　　处：《信息技术》2024年第11期105-111,119,共8页Information Technology

基　　金：2020年上海市上海大学横向科研项目(19H01488)。

摘　　要：随着互联网视频图像数据资源的日益增加,以及机器学习的算力与技术的不断突破,海量图文影像的价值特征提取具有可行性。在艺术图像领域,图像的色彩、肌理和形状所蕴含的情感语义特征价值尤为重要。现今的图像情感特征研究领域,使用卷积神经网络模型对图像情感语义进行训练的较多,但关于艺术图像多层次特征感知方向的探究甚少。基于此,文中提出融合注意力机制的多特征融合模型,以提取图像的底层特征,使用融入混合注意力机制的ResNet101模型提取图像深层情感语义特征,实现对艺术图像的情感语义识别。With the increasing of Internet video image data resources and the continuous breakthrough of machine learning computing power and technology,it is feasible to extract the value features of massive graphic images.In the field of artistic images,the value of the color,texture and shape of the image is particularly important.Nowadays,in the field of image emotion feature research,the convolutional neural network model is often used to train image emotion semantic,but the multi-level feature perception direction of artistic images is rarely explored.Based on this,this paper proposes a multi-feature fusion model integrating attention mechanism to extract the underlying features of images,and uses ResNet101 model integrating mixed attention mechanism to extract the deep emotional semantic features of images to realize emotional semantic recognition of artistic images.

关 键 词：特征融合 视觉感知 注意力机制 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]