基于卷积神经网络和SVM的中国画情感分类

发布时间：2018-05-26 10:53

  本文选题：图像情感 + 中国画　； 参考：《南京师大学报(自然科学版)》2017年03期

【摘要】：图像情感是指计算机识别数字图像所表达内容引起人的情感反应,根据不同的情感反应,可以对不同的图像进行分类.在信息量急剧增长的今天,图像情感分类有助于图像的标注和检索,蕴藏着很大的社会和商业价值.不同于西洋画的"以形写形",中国画有着自己明显的特征:传统的国画不讲焦点透视,不强调自然界对于物体的光色变化,不拘泥于物体外表的肖似,而多强调抒发作者的主观情趣.这比弥合一般的低层特征和人类情感高层语义之间的鸿沟的难度更大.基于卷积神经网络因为其具有结构简单、适应性强、训练参数少、连接点多等特点,可以直接输入原始图像,能够避免对图像进行复杂的前期预处理.相比传统图像特征提取方法,卷积神经网络具有明显的优势.本文的目的是利用卷积神经网络发掘低层特征和情感语义之间的联系,提取国画图像特征,对得到的特征进行PCA降维、归一化等操作后,利用支持向量机(SVM)分类器进行情感分类.
[Abstract]:Image emotion refers to the human emotional response caused by computer recognition of digital image expression. According to different emotional reactions, different images can be classified. With the rapid increase of information, image emotional classification is helpful to image tagging and retrieval, and it contains great social and commercial value. Unlike Western paintings, Chinese painting has its own obvious characteristics: traditional Chinese painting does not stress focus perspective, does not emphasize the natural changes in the light and color of objects, and does not stick to the appearance of objects. And more emphasis on the subjective feelings of the lyricist. This is more difficult than bridging the gap between the general low-level features and the high-level semantics of human emotions. Based on convolution neural network, because of its simple structure, strong adaptability, less training parameters and more connecting points, it can directly input the original image and avoid the complicated pre-processing of the image. Compared with the traditional image feature extraction method, the convolution neural network has obvious advantages. The purpose of this paper is to use convolution neural network to explore the relationship between low-level features and emotional semantics, extract the features of traditional Chinese painting image, and perform PCA dimensionality reduction, normalization and other operations. Support vector machine (SVM) classifier is used to classify emotion.
【作者单位】： 大数据分析与系统实验室(天津大学软件学院);天津大学计算机科学与技术学院;
【基金】：国家自然科学基金(61572351;61772360)
【分类号】：J212;TP18;TP391.41
，

本文编号：1936963