结构基元可辨尺度下准周期织物纹理的表征、解耦及特征提取研究

项目名称： 结构基元可辨尺度下准周期织物纹理的表征、解耦及特征提取研究

项目编号： No.61202310

项目类型： 青年科学基金项目

立项/批准年度： 2013

项目学科： 计算机科学学科

项目作者： 潘如如

作者单位： 江南大学

项目金额： 25万元

中文摘要： 机织物图像是一种典型准周期纹理，其相关检测与分析，如疵点检测、毛球评级等都是基于纹理分析进行。由于缺乏对准周期纹理基础构成的深入研究和理解，现有解决方法适用范围窄、通用性差，算法有待突破。本项目研究在结构基元可辨尺度下准周期织物纹理的表征、解耦及特征提取方法，以得到通用性的织物检测与分析方法。首先模拟准周期织物纹理，利用时频变换方法在时域和频域内建立织物纹理的表征模型；然后基于纹理表征模型，确定神经元数目及权重系数，进而构建Gabor小波神经网络对准周期耦合纹理进行解耦，实现织物结构背景纹理与检测目标纹理的分离；最后利用纹理分割方法定位检测目标，提取检测目标的统计、灰度及分布特征，完成织物的检测与分析。项目研究将得到模块化的准周期织物纹理表征、解耦及特征提取方法，为智能化的织物及准周期纹理的检测与分析奠定基础，研究成果有助于深化信息检测理论在纺织工业中应用，同时提供准周期纹理分析示范方案。

中文关键词： 准周期织物纹理；结构基元；纹理解耦；纹理表征；特征提取

英文摘要： Woven fabric image is a kind of quasi-periodic texture and its detection and analysis, such as defect detection, pilling grading, are all processed based on texture analysis. As the lack of deeply research and analysis of structure primitive of quasi-periodic texture, the current solution methods have the drawbacks of narrow scope of application and poor universality. The current algorithms remain to be broken through. This project is to characterize and decouple the quasi-periodic fabric texture, extract the features from fabric texture in the discernible dimension of structure primitive. The universality methods used for fabric detection and analysis will be obtained. First, the quasi-periodic texture is simulated and time-frequency transform method is used to construct the characterization model of fabric texture in time and frequency domain. Then, the number of neurons and the weighting coefficient are determined by the characterization model of quasi-periodic texture. Gabor wavelet neural network will be constructed to decouple the quasi-periodic coupling texture. The backgroud texture of fabric structure and the target texture will be seperated successfully. Last, the target objects will be detected and located with texture segmentation method. The statistical features, gray features and distribution featu

英文关键词： quasi-periodic fabric texture；structure primitive；texture decoupling；texture characterization；feature extraction