面向动漫素材的特征提取与分类识别

发布时间：2018-06-25 15:15

本文选题：空间金字塔匹配 + 上下文相关直方图　；参考：《浙江大学》2011年硕士论文

【摘要】：从全球来看,动漫产业已经成为一个庞大的产业。然而在我国动漫产业还是一个新兴产业,起步不久但发展迅速。在动漫产业飞速发展的背景下,面对海量的数字化动漫素材,如何结合动漫素材的视觉特点,有效地提取图像特征,并在此基础上实现高效的分类识别成为亟待解决的问题。因此,利用数字图像处理技术解决上述问题成为本文的研究动机。动漫素材,主要以图像的形式存在。剪纸图像,作为一种典型的动漫素材,在图像特征和人类视觉特性上都具有不同于普通图像的特点。因此,本文的研究目标是：以剪纸图像为主要研究对象,根据剪纸图像上述特性,对基于剪纸图像的特征提取与分类识别进行研究,进而为上述问题提出了解决方案。本文首先针对中国剪纸识别中存在底层形状特征难以表达高层语义这一“语义鸿沟”问题,提出的基于空间约束特征组合与选择的中国剪纸识别算法将空间金字塔匹配和上下文相关直方图这两种图像特征提取方法结合起来,有效地克服了其在表达图像形状上的局限性,并在实验中验证其有效性。然后针对传统图像分类识别框架由于在特征提取阶段没有引入图像空间信息,导致图像特征表达能力不足,从而制约了分类识别正确率的提高这一问题,在基于空间信息的中国剪纸特征提取方法的基础之上,进一步提出了基于特征选择与组合的中国剪纸分类识别方法。最后通过实验对比和分析验证了基于特征选择与组合的中国剪纸分类识别算法的有效性。
[Abstract]:From the global perspective, animation industry has become a huge industry. However, the animation industry in China is still a new industry, starting soon but rapid development. Under the background of the rapid development of animation industry, how to combine the visual characteristics of animation material, how to extract image features effectively, and how to achieve efficient classification and recognition becomes a problem to be solved urgently in the face of mass digital animation material. Therefore, the use of digital image processing technology to solve the above problems has become the motivation of this paper. Animation material, mainly in the form of images. As a typical animation material, paper-cut image has different features from ordinary images in image features and human visual characteristics. Therefore, the research goal of this paper is: take paper-cut image as the main research object, according to the above characteristics of paper-cut image, study the feature extraction and classification recognition based on paper-cut image, and then put forward the solution to the above problems. In this paper, we first aim at the problem of "semantic gap" in Chinese paper-cut recognition, in which the underlying shape features are difficult to express high-level semantics. The proposed Chinese paper-cut recognition algorithm based on spatial constraint feature combination and selection combines spatial pyramid matching and context-dependent histogram as two image feature extraction methods. It overcomes the limitation of image shape expression and proves its validity in experiments. Then because the traditional image classification recognition framework does not introduce the image spatial information in the feature extraction stage, the ability of image feature expression is insufficient, which restricts the improvement of classification recognition accuracy. Based on the feature extraction method of Chinese paper-cut based on spatial information, a Chinese paper-cut classification and recognition method based on feature selection and combination is proposed. Finally, the effectiveness of the Chinese paper-cut classification recognition algorithm based on feature selection and combination is verified by experimental comparison and analysis.
【学位授予单位】：浙江大学
【学位级别】：硕士
【学位授予年份】：2011
【分类号】：TP391.41

【参考文献】