剪纸纹样识别算法研究

发布时间：2018-09-07 07:34

【摘要】： 图象识别技术应用领域十分广泛,是当前研究的热点问题之一,很多的专家学者都在这方面进行了研究,取得了一些很好的成果。剪纸是我国历史悠久的传统民间艺术之一,随着动漫产业的发展,剪纸艺术作品是一种很好的动漫素材。由于艺术作品变形大,对艺术作品图象的识别研究较少,将剪纸艺术与图象识别结合起来,研究计算机剪纸将是一项非常有意义的工作。特征提取是决定相似性和识别图象的关键,如何提取到图象本质的且具有不变的有效特征是研究核心内容。本文深入研究了国内外图象识别的各种特征提取方法和识别方法的研究现状,针对非数学变换的变形图象识别,提出了一些较为有效的方法,能够提取到不受平移、旋转、尺度变化和小变形的有效特征,并应用于剪纸纹样的艺术图象识别中,实验证明有良好的效果。论文的工作主要从以下几个方面展开: (1)研究剪纸图象的特点,对剪纸纹样进行总结和分类。使用一般的图象预处理技术,通过背景去噪、灰度化以及二值化处理等来处理剪纸图象,能够有效的去除剪纸图象的背景噪声,突出了纹样,为后续的纹样识别工作做准备。 (2)提出一种R-变换与奇异值分解的特征提取算法,本文在radon变换的基础上,提出了一种较为简单的方法进行图象特征提取。使用该方法提取到的特征具有平移、旋转和尺度不变性,且具有一定的鲁棒性,代表了图象的结构特征,能够较好的识别一定变形剪纸纹样。 (3)针对现有的方法提取到的特征不适合变形图象的缺点,提出一种适合剪纸纹样的识别方法。利用Fourier-Mellin变换具有不受几何变换的结果,使用小波分析求其各层的方差和均值,得到目标不同子带的特征值。实验证明该特征不仅具有平移、旋转和尺度不变性,而且适用于有变形的纹样图象的识别。 (4)研究各种图象识别的方法,使用分类性能和泛化能力较好的支持向量机作为纹样识别的分类器,能够将纹样有效进行识别和分类。实验所用的剪纸图象均为扫描剪纸相关书籍得到,通过实验对论文中的算法进行了理论分析和验证,结果表明: (1)基于R-变换和奇异值分解的方法,计算简单,提取到的特征鲁棒性好,具有平移、旋转和尺度不变性,对多数有变形的图象有较好的区分度; (2)多分辨FM变换算法不受几何变换影响,能有效地实现剪纸纹样识别与分类,并具有较好的鲁棒性; (3)使用支持向量机作为分类器,具有较好的泛化能力,能够将夸张变形的纹样有效的分开; 本论文从实际问题出发,深入研究了剪纸纹样图象的预处理、特征提取和识别方法等技术,针对非严格意义上的数学变形图象,提出一些特征提取和识别的有效办法。本文算法提取到的特征不但具有平移、旋转和尺度等几何不变性,而且能够进行一定的变形图象的识别,在理论上拓宽了图象识别的方法。结合剪纸艺术进行纹样识别研究,为计算机剪纸艺术的设计与实现提供了一种新方法。
[Abstract]:The application of image recognition technology is very extensive, and it is one of the hotspots of current research. Many experts and scholars have done research in this field and have made some good achievements. Paper-cut is one of the traditional folk arts with a long history in China. With the development of animation industry, paper-cut is a good material for animation. It will be a very meaningful work to combine the paper-cut art with the image recognition to study the computer paper-cut.
Feature extraction is the key to determining similarity and recognizing an image.How to extract the essential and invariant effective features of an image is the core content of the research.In this paper,the research status of various feature extraction and recognition methods for image recognition at home and abroad is deeply studied,and some comparisons are put forward for deformed image recognition based on non-mathematical transformation. For the sake of effective method, it can extract the effective features without translation, rotation, scale change and small deformation, and is applied to the recognition of the artistic image of paper-cut pattern. The experiment proves that the method has a good effect.
The work of this paper is mainly from the following aspects:
(1) Studying the characteristics of the paper-cut image, summarizing and classifying the pattern of paper-cut. Using the general image pretreatment technology, the paper-cut image can be processed by background denoising, graying and binary processing, which can effectively remove the background noise of the paper-cut image, highlight the pattern and prepare for the follow-up pattern recognition.
(2) A new feature extraction algorithm based on R-transform and singular value decomposition is proposed in this paper. Based on radon transform, a simpler method is proposed to extract image features. The features extracted by this method are invariant in translation, rotation and scale, and have certain robustness. It represents the structural features of the image and can be better. Recognition of certain deformation paper cut patterns.
(3) To overcome the disadvantage that the features extracted by existing methods are not suitable for deformed images, a paper-cut pattern recognition method is proposed. Fourier-Mellin transform is used to obtain the eigenvalues of different subbands of the target by calculating the variance and mean of each layer of the feature without geometric transformation. Translation, rotation and scale invariance, and suitable for recognition of deformed patterns.
(4) Studying all kinds of image recognition methods, using the support vector machine with better classification performance and generalization ability as the classifier of pattern recognition can effectively recognize and classify the pattern.
The paper-cut images used in the experiment are all obtained by scanning paper-cut related books. The algorithm in the paper is theoretically analyzed and verified by experiments. The results show that:
(1) Based on R-transform and singular value decomposition, the method is simple in calculation, robust in feature extraction, invariant in translation, rotation and scale, and can distinguish most deformed images.
(2) Multi-resolution FM transform algorithm is not affected by geometric transformation, and can effectively recognize and classify paper-cut patterns with good robustness.
(3) Support Vector Machine (SVM) is used as classifier, which has good generalization ability and can effectively separate exaggerated and deformed patterns.
Starting from practical problems, this paper deeply studies the pretreatment, feature extraction and recognition methods of paper-cut pattern images, and puts forward some effective methods of feature extraction and recognition for non-strictly mathematically deformed images. It can recognize certain deformed images, which broadens the method of image recognition in theory. It provides a new method for the design and Realization of computer paper-cut art by combining the paper-cut art to study pattern recognition.
【学位授予单位】：广西师范大学
【学位级别】：硕士
【学位授予年份】：2008
【分类号】：TP391.41

【相似文献】