Computer Science and Application
Vol.4 No.03(2014), Article ID:13293,6 pages
DOI:10.12677/CSA.2014.43011

Classification of Landscape Painting Texturing Based on Gabor

Yufan Li, Hongyan Xing, Jingxuan Chen, Minzhi Yang

School of Applied Mathematics, Guangdong University of Technology, Guangzhou

Email: justin_yufan@163.com, 928hongy@163.com, 1054637410@qq.com, 593097322@qq.com

Received: Feb. 2nd, 2014; revised: Mar. 3rd, 2014; accepted: Mar. 12th, 2014

ABSTRACT

Extracting the effective features for texture description and classification has always been the hot spot of the texture analysis. In this paper, according to different texture of traditional Chinese painting, we use a kind of Gabor filter technique to classify the painting. By texture feature extraction, first of all, we preprocess the traditional Chinese painting images with geometric normalization and light normalization, after that we process the group of the Gabor filter of high dimensional feature vectors by principal component analysis (PCA) for dimension reduction. Finally, support vector machine (SVM) method is employed for texture classification. The accuracy rate of this classification method can reach 95.5%.

Keywords:Texturing Classification; Gabor Filter; Principal Component Analysis; Support Vector Machine

1. 引言

2. 系统设计

3. 预处理

4. 国画的纹理特征提取

Figure 1. Structure chart

(a) 原图                  (b) 裁剪图                 (c) 均衡化

Figure 2. Preprocessing an instance of the image

1) 就最小化空间域和频率域的联合二维不确定来讲，Gabor小波是最优的。Gabor小波表示了这样粗糙时，空间域采样范围应较大，频率域采样范围应较小。

2) Gabor小波的方向和尺度可调[1] 。

3) Gabor滤波器可以被视为方向和尺度均可变化的边缘和直线(条纹)的检测器，且对于一个给定区域中微观特征的统计，通常可用来表示基本的纹理信息。

4.1. Gabor小波

Gabor小波的核函数定义如下：

(4.1.1)

(4.1.2)

4.2. 利用Gabor小波进行特征提取

(4.2.1)

(a)(b)

Figure 3. Diagram of the Gabor wavelet real component when. (a) The waveform of Gabor kernel function real part; (b) The waveform of Gabor kernel function imaginary part

Figure 4. Diagram of the Gabor wavelet real component

(4.2.2)

(4.2.3)

5. 国画特征降维及分类

Figure 5. The traditional Chinese painting in the Gabor texture feature of image

1963年，Vapnik在解决模式识别问题时提出了支持向量方法，这种方法从训练集中选择一组特征子集，使得对特征子集的划分等价于对整个数据集的划分，这组特征子集就被称为支持向量(SV)。SVM考虑寻找一个满足分类要求的超平面，并且使训练集中的点距离分类面尽可能地远，也就是寻找一个分类面使它两侧的空白区域(margin)最大。SVM在模式识别、回归函数估计、预测等大量应用中取得了良好的效果。

6. 实验结果与分析

6.1. 支持向量机参数寻优方法[5]

6.2. SVM的分类结果

Table 1. The accuracy comparison of different search methods of Sonar data sets

Figure 6. The results of SVC parameter selection (3D)

Figure 7. The classification results of the some texture of point, line and plane

Figure 8. The classification contrast of SVM, BP neural network and discriminant analysis

6.3. SVM、BP神经网络和判别分析分类准确率对比

7. 总结

1. [1]   朱希安, 曹林, 编著 (2012) 小波分析及其在数字图像处理中的应用. 电子工业出版社, 北京.

2. [2]   邓洪波, 金连文 (2007) 一种基于局部Gabor滤波器组及PCA + LDA的人脸表情识别方法. 中国图象图形学报, 2, 322-329.

3. [3]   陈洋, 王润生 (2007) 结合Gabor滤波器和ICA技术的纹理分类方法. 电子学报, 2, 299-303.

4. [4]   Chang, C.C. and Lin, C.J. (2011) LIBSVM: A library for support ectormachines.
http://www.csie.ntu.edu.tw/~cjlin/libsvm/

5. [5]   林升梁, 刘志 (2007) 基于RBF核函数的支持向量机参数选择. 浙江工业大学学报, 2, 163-167.

6. [6]   王健峰, 张磊, 陈国兴, 等 (2012) 基于改进的网格搜索法的SVM参数优化. 应用科技, 3, 28-31.