纹理为物体的材料特性及种类识别提供非常重要的信息。目前纹理识别研究主要依托于无混杂情况的物体表面材料的实例分类，通过提取边缘、点或条状等局部低级特征来表征纹理。这种方式越来越不能满足一些实际应用对纹理识别方法的需求，例如纹理语义检索等基于可描述感知特征的纹理分类等。这就必须考虑纹理周期性、规律性，方向性及结构复杂程度等高级感知特征在描述纹理方面的作用。本项目针对规律或近规律纹理研究其高级感知特征的表征和计算，包括对纹理基元形状及空间排布、纹理方向性、基元密度及占空比进行量化定义和检测，并验证其在不同纹理识别应用上的有效性，而且针对实际交通拥堵中车辆队列形成的近规律纹理或户外植被形成的方向性纹理的快速有效检测以达到预警或问题分析。本项目在探索纹理识别理论分析与实际应用相互促进、展开纹理分析识别研究新局面具有重要意义。

Texture provides useful cues of material properties of objects and their identity. Research in texture recognition often concentrates on the problem of material recognition in uncluttered conditions by extracting the local low-level features such as edges, points or bars. The methods formulated and developed under this background cannot be directly and effectively used in semantic texture retrieval, real-world material segmentation and other tasks, where high-level perceptual features such as the periodicity, regularity, directionality and complexity of the texture are necessary. We propose to investigate the computational methods for high-level structural features including texton and their spatial arrangement, texture direction and texton density etc. We will not only verify their usefulness on different texture recognition applications, but also apply them to detect real traffic congestion by considering the vehicle queues as nearly regular texture, and detect outdoor directional vegetation. This project has great significance in promoting mutually the theory research and real applications of texture recognition, and providing new ideas for texture analysis and recognition.