基于双树复小波变换和邻域操作的哈密瓜纹理提取

Texture Extraction of Hami Melon Based on Dual-tree Complex Wavelet Transform and Neighborhood Operation

为研究哈密瓜表面纹理特征分布规律,采集金密16号9成熟、全熟和金密17号9成熟、全熟共168幅哈密瓜样本图像,对RGB彩色图像的R、G、B分量执行代数运算,转换为灰度图后进行背景分割,然后利用双树复小波变换(DT-CWT)分解图像,获取高频子图像,并对其执行邻域操作,采用迭代法选取最优阈值完成纹理提取,最后利用灰度差分统计法和纹理频谱分析法描述分析哈密瓜纹理特征,建立基于支持向量机(SVM)的分类模型。研究结果表明,利用DT-CWT和邻域操作相结合的方法可得到更加连续、完整的哈密瓜纹理图像;4种哈密瓜的纹理特征值差异显著,利用纹理特征值分类准确率为89.3%;哈密瓜表面纹理无周期性。

In order to investigate the distribution feature of surface texture,168 images of Hami melon samples from two different varieties in two kinds of ripeness were acquired. The algebra operations were conducted in terms of R,G,B components,and the gray images were obtained to implement the background segmentation. Then,the images were decomposed by dual-tree complex wavelet transform（ DT-CWT） to obtain high frequency sub-images. Following the neighborhood operation,the extraction results were derived from selecting the optimal thresholds by iterative method. Finally,the methods of gray-scale differential statistics and texture frequency analysis were used to analyze the texture feature,support vector machine（ SVM） was employed to build a model for texture classification. Results of computer simulation indicated that more continuous and complete images were obtained when DT-CWT and image neighborhood operation were employed to extract texture. There were significant differences among texture eigenvalues of four types of Hami melons,and the accuracy rate of classification was89. 3%. In addition,periodic characteristic was not found from the appearance texture.