摘要
油画分类是油画生成、油画识别及数字油画应用的重要基础工作。但由于油画图片与普通图片存在较大的质感差异,而且是油画家的个性化创作,不确定性更高,较普通照片的分类更困难。论文以分类出含有桥梁的油画为例,提出一种基于近邻卷积神经网络的油画分类方法,利用K最近邻分类算法提取与测试样本最接近的K个训练样本,卷积神经网络挖掘油画中的深层特征,从而对油画中的对象进行分类。论文详细讨论了数据处理、卷积神经网络的架构设计、训练过程。并在kaggle数据集上对该方法进行了分析与比较,使用三个数据集进行实验,实验结果表明该方法较最近邻算法精度上平均提高了2.4%,较卷积神经网络精度上平均提高了3.1%,较支持向量机方法精度上平均提高了6.9%。
The classification of oil paintings is important for the generation,recognition and application of oil paintings.However,due to the large difference in texture between oil paintings and ordinary pictures,and the personalized creation of oil painters,and the higher uncertainty,it becomes more difficult to classify oil paintings.Taking the classification of oil paintings containing bridges as an example,this paper proposed a method for oil painting classification based on the nearest-neighbor convolutional neural network.Utilizing the K-nearest neighbor(KNN)algorithm to extract the K closest training samples,we investigated the deep features of the oil paintings with the convolutional neural network and distinguished the objects in them.We discussed in detail the data processing,architecture design and training process of convolutional neural networks.Through the analysis and comparison of this method on kaggle dataset,we selected three datasets for experiments.The results show that this method has improved the accuracy by an average of 2.4%compared to the nearest neighbor algorithm,3.1%compared to the convolutional neural network,and 6.9%compared to the support vector machine method.
作者
钱华
祁枢杰
顾涔
陶然
吴宏杰
QIAN Hua;QI Shujie;GU Cen;TAO Ran;WU Hongjie(Suzhou Art&Design Technology Institute,Suzhou 215104,China;School of Electronic&Information Engineering,SUST,Suzhou 215009,China)
出处
《苏州科技大学学报（自然科学版）》
CAS
2024年第1期69-75,共7页
Journal of Suzhou University of Science and Technology（Natural Science Edition）
基金
国家自然科学基金项目(62372318,62073231,62176175)
国家重点研发计划项目(2020YFC2006602)
苏州大学江苏省计算机信息处理技术重点实验室项目(KJS2166)
苏州大学江苏省大数据智能工程实验室开放课题(SDGC2157)。
关键词
卷积神经网络
K最近邻分类算法
数据可视化
图像分类
convolutional neural network
K-nearest neighbor
data visualization
image classification
作者简介
钱华(1982-),女,江苏苏州人,讲师,硕士,研究方向:实时动画,图形图像处理,数字媒体,E-mail:13854184@qq.com。
