Abstract: Texture classification has long been an important research topic in image processing.Nowadays texture classification based on mathematical parameters is very popular，but fails to break through the semantic obstacle between visual features and human understanding of textures.In this paper，a novel approach of texture classification based on conceptual words of Chinese natural language which describe various natural textures has been put forward.Ten texture classes are presented to automatically classify natural texture with wavelet packet and LSSVM，which transform texture visual features to semantic description.Experimental results show that this approach is useful to negotiate the “semantic gaps” between texture concepts and feature parameters on image understanding and image retrieval based on natural language.

Key words: texture classification, wavelet transform, LSSVM

摘要： 纹理分类一直是图像处理领域重要的研究课题之一。目前，用数学方法描述纹理特征从而进行纹理分类非常流行，但这些方法无法消除纹理视觉特征和人们理解的纹理概念之间的语义障碍。提出了一种新的基于中文自然语言纹理描述词的纹理方法，把常见的自然纹理分为10大类别，然后利用小波包分解和最小二乘支持向量机对自然纹理进行分类，实现了纹理的视觉特征到语义描述的转换。实验结果证明，该方法在图像理解和基于自然语言的图像检索中有助于缩小纹理特征的数学描述和人类理解之间的“语义鸿沟”。

关键词: 纹理分类, 小波分析, LSSVM