基于元胞自动机和特征加权的花卉图像分类

发布时间：2018-02-01 16:28

本文关键词： 图像分割元胞自动机特征融合特征加权花卉图像分类　出处：《太原科技大学》2017年硕士论文　论文类型：学位论文

【摘要】：随着人们生活质量的提高,养植花卉成为许多人培养性情的一种爱好。面对如此种类繁多、色彩纷杂的花卉,研究一种有效的花卉图像分类方法,帮助人们更好地认识花卉的一般形态和生活习性,提高花卉养植水平,具有重要的研究价值。图像分割和特征融合是提高花卉图像分类精度的两个主要步骤,但是传统的图像分割方法常常会因花卉图像背景过于复杂而造成分割效果不佳,而且一般的特征融合方法也仅是简单地把多个特征拼接在一起,并未将不同特征对花卉分类贡献的不同考虑在内,从而影响了分类的效果。为进一步提高花卉图像分类的精度,本文提出一种基于元胞自动机和特征加权融合的花卉图像分类方法。主要研究内容包括以下三个方面:(1)给出了一种基于元胞自动机的花卉主体区域提取方法。该方法首先对花卉图像进行预处理操作,应用SLIC算法将其分割成N个小的超像素点,并通过和分类的边缘种子进行颜色空间和距离空间的对比以得到一幅基于背景的显著图。其次,根据相应的规则,在基于背景的显著图上使用元胞自动机的新型传播机制得到优化的显著图。然后,再采用最大类间方差法找到该显著图的一个合适的阈值,以完成灰度图转换为二值图的操作。最后,在花卉原图的基础上将二值图的白色部分进行填充,得到了花卉主体区域。通过在17-flower数据集上进行实验,验证了该方法是有效的。(2)给出了一种基于花卉主体区域的特征加权融合分类方法。一般的特征融合方法仅是简单地将多个特征拼接在一起,并未把不同特征对花卉分类贡献的不同考虑在内。为有效提高花卉图像分类精度,本文首先对上述(1)方法中提取的花卉主体区域的颜色特征和局部特征进行加权融合,然后利用SVM实现了花卉图像分类。最后,通过在13-flower和102-flower图像数据集上进行实验,验证了该方法的有效性。(3)开发了一个基于元胞自动机和特征加权融合的花卉图像分类原型系统。以Matlab7.0和VB6.0作为开发工具,设计并实现了一个基于元胞自动机和特征加权融合的花卉图像分类原型系统。
[Abstract]:With the improvement of people's quality of life, planting flowers has become a hobby for many people to cultivate temperament. In the face of so many kinds of flowers and colorful flowers, an effective flower image classification method is studied. It has important research value to help people better understand the general form and life habits of flowers and improve the level of flower planting. Image segmentation and feature fusion are the two main steps to improve the classification accuracy of flower images. However, traditional image segmentation methods often result in poor segmentation results due to the complexity of flower image background, and the general feature fusion method is only simple to combine multiple features together. The contribution of different characteristics to flower classification is not taken into account, thus affecting the effect of classification, in order to further improve the accuracy of flower image classification. In this paper, a flower image classification method based on cellular automata and feature weighted fusion is proposed. In this paper, a method of extracting flower body area based on cellular automata is presented. Firstly, the preprocessing operation of flower image is carried out. Using SLIC algorithm, it is divided into N small super-pixel points, and the color space and the distance space are compared with the classified edge seeds to obtain a significant map based on background. According to the corresponding rules, the new propagation mechanism of cellular automata is used to obtain the optimized salience graph on the background salient graph. Then, a suitable threshold value of the significant graph is found by using the maximum inter-class variance method. Finally, the white part of the binary image is filled on the basis of the original flower map. The main region of flower was obtained. The experiment was carried out on the 17-flower data set. It is proved that this method is effective. (2) A new method of feature weighted fusion classification based on flower body region is presented. The general feature fusion method is only simple to join several features together. The contribution of different characteristics to flower classification is not taken into account in order to improve the classification accuracy of flower image effectively. In this paper, the color features and local features of the flower body area extracted from the above method are firstly weighted fusion, and then the flower image classification is realized by using SVM. Finally. Experiments were carried out on 13-flower and 102-flower image datasets. The validity of this method is verified. (3) A flower image classification prototype system based on cellular automata and feature weighted fusion is developed. Matlab7.0 and VB6.0 are used as development tools. A flower image classification prototype system based on cellular automata and feature weighted fusion is designed and implemented.
【学位授予单位】：太原科技大学
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：S68;TP391.41

【参考文献】