摘要
作为人类精神活动产物的艺术图像,其本身蕴含着丰富的情感语义信息,研究艺术图像的情感分类有助于艺术图像的鉴赏与保护,以图像为对象的情感分类研究已成为情感计算的研究热点,但该分类主要依赖于图像低层特征的抽取,从而导致图像情感分类结果不高。本文提出了一种基于底层特征和注意力机制的艺术图像情感分类模型,即提取艺术图像的CLAHE颜色特征、Laplacian纹理特征与艺术图像深度特征融合作为输入,同时引入CBAM注意力机制以关注图像的重点区域,构建面向艺术图像情感分类的卷积神经网络模型FeatursNet。实验结果表明,应用本文提出的模型在艺术图像数据集上的情感分类准确率可达93.36%,相较于其他模型有较大提升。
Emotion classification based on image has become a research hotspot of emotion computing.However,the classification of image emotion mainly depends on the extraction of deep features,which leads to low classification results.As the product of human spiritual activities,art images itself contains rich emotional semantic information.Research on the emotional classification of art images is helpful to the appreciation and protection of art images.An emotion classification model of art images based on deep-level features and attention mechanism network is proposed in this paper which extracts the CLAHE color features and Laplacian texture features and fuses them into their deep-level features of art images as inputs.At the same time,an attention mechanism based on CBAM is introduced to pay close attention to important areas,and a model named as FeaturesNet is constructed for classifying emotion of art images.The experimental results show that the emotion classification accuracy of this improved convolutional neural network model on the image data set of art images defined in this paper can reach 93.36%,which is greatly improved compared with other models.
作者
杨松
刘佳欣
吴桐
YANG Song;LIU Jiaxin;WU Tong(School of Software,Dalian University of Foreign Languages,Dalian Liaoning 116044,China;Research Center for Language Intelligence,Dalian University of Foreign Languages,Dalian Liaoning 116044,China;Research Center for Networks Space Multi-Languages Big Data Intelligence Analysis,Dalian University of Foreign Languages,Dalian 116044,China)
出处
《智能计算机与应用》
2022年第2期126-132,共7页
Intelligent Computer and Applications
基金
国家自然科学基金(61806038)
辽宁省社会科学规划基金(L18BTQ005)
辽宁省教育厅科学研究项目(2019JYT07)
关键词
艺术图像
情感分类
卷积神经网络
深度特征
注意力机制
artistic images
emotional classification
convolutional neural network
deep features
attention mechanism
