基于视觉显著性的织物疵点检测算法研究

发布时间：2018-03-05 13:17

本文选题：织物疵点　切入点：疵点检测　出处：《中原工学院》2017年硕士论文　论文类型：学位论文

【摘要】：织物疵点严重影响织物质量,对织物疵点的有效自动检测已成为纺织品质量评价的关键之一。由于织物疵点图像纹理复杂多变,给基于机器视觉的疵点自动检测带来了挑战。近年来,人们模拟生物的视觉注意机制引入了基于视觉显著性的图像处理和模式识别检测方法,并取得了良好的效果。本文对基于视觉显著性的织物疵点检测算法进行了深入探究。针对织物图像复杂纹理特点,考虑其方向性和随机性,模拟人类视觉感知通路的层次化处理机制,构建了基于视觉显著性的织物疵点检测模型,提出了有效的织物疵点检测算法。采用小波变换或平稳小波变换与背景估计相结合,提出了一种基于平稳小波变换和背景估计的织物疵点检测算法。首先,分别对织物图像进行小波变换或平稳小波变换获得特征图像;其次,采用分区处理得到特征图像的多个背景图,求取原始图像与每一个背景图的欧氏距离生成多个子显著图,经融合后获得包含候选疵点区域的视觉显著图;最后,基于全局估计的高斯分布模型获得疵点显著图,经图像分割检测出疵点。实验结果表明,该算法能准确定位疵点区域,实现对织物疵点的有效检测且平稳小波变换具有更高的检测准确率。利用互信息具有的基本性质,提出了一种基于互信息测度及上下文分析的织物疵点检测算法。首先,对织物图像进行均匀的部分重叠的分块处理获得图像块;其次,分别计算每个图像块与周围的K个图像块两两之间的信息熵;最后,基于上下文分析的图像块信息熵比对获得视觉显著图,经阈值分割检测出疵点。实验结果表明,该算法无需特征提取就可清晰定位疵点区域,实现对织物疵点的有效检测且优于现有的相似性检测算法的检测结果。
[Abstract]:Fabric defects seriously affect fabric quality. Effective automatic detection of fabric defects has become one of the key issues in textile quality evaluation. In recent years, the visual attention mechanism of simulated organisms has introduced visual salience based image processing and pattern recognition detection methods. In this paper, the fabric defect detection algorithm based on visual salience is deeply explored. Considering the complex texture characteristics of fabric image, the direction and randomness of fabric defect detection algorithm are considered. This paper simulates the hierarchical processing mechanism of human visual perception path, constructs a fabric defect detection model based on visual salience, and proposes an effective fabric defect detection algorithm, which combines wavelet transform or stationary wavelet transform with background estimation. A fabric defect detection algorithm based on stationary wavelet transform and background estimation is proposed. Firstly, the fabric image is obtained by wavelet transform or stationary wavelet transform. The multi-background images of feature images are obtained by partition processing, and the Euclidean distance between the original image and each background image is obtained. After fusion, the visual salience images containing candidate defect regions are obtained. Finally, the Euclidean distance between the original image and each background image is obtained. Gao Si distribution model based on global estimation obtains defect salience map and detects defects by image segmentation. Experimental results show that the proposed algorithm can locate defect region accurately. In order to realize the effective detection of fabric defects and the higher detection accuracy of stationary wavelet transform, a fabric defect detection algorithm based on mutual information measurement and context analysis is proposed. The fabric image is partitioned evenly and partially overlapped to obtain the image block. Secondly, the information entropy between each image block and the surrounding K image blocks is calculated respectively. Finally, The information entropy of image blocks based on context analysis is used to obtain visual saliency images, and defects are detected by threshold segmentation. Experimental results show that the proposed algorithm can clearly locate defect regions without feature extraction. The detection results of fabric defects are better than the existing similarity detection algorithms.
【学位授予单位】：中原工学院
【学位级别】：硕士
【学位授予年份】：2017
【分类号】：TS101.97;TP391.41

【参考文献】