摘要
为提高织物疵点检测率,将Gabor滤波法与等距映射方法进行融合,克服疵点检测过程中存在的问题。首先用由3个尺度和5个方向组成的15个Gabor滤波器簇对织物疵点图像进行滤波,减少疵点图像光照不均和对比度低的影响;然后将滤波图像划分成面积相等且互不重合的邻域,并从邻域中提取高维特征向量。采用等距映射方法对高维特征向量进行降维,剔除高维特征中冗余信息,强化分类器拟合能力;再用低维嵌入模型提取新增样本低维特征向量,用于概率神经网络分类器分类,检测是否存在疵点;最后用2种不同纹理的织物进行检测实验。结果表明,本文方法能有效提高疵点的检测精度。
In order to improve the correction rate of fabric defects,Gabor filters and Isomap were used to detect the fabric defect. Firstly,the images of fabric defect were filtered by 15 Gabor filters with 3orientations and 5 scales,which contributed to overcome the effect of uneven illumination and low contrast. Then,the filtered images were divided into non-overlapping rectangular patches and highdimensional features were extracted. Simultaneously, Isomap algorithm was applied to reduce the dimensionality of feature and eliminate the redundant information. Besides,a mapping model of new samples was proposed to detect the low dimensional embedding results. Finally,the performance of proposed algorithm was estimated off-line by two sets of fabric defect images. The theoretical argument is supported by experimental results.
出处
《纺织学报》
EI
CAS
CSCD
北大核心
2017年第3期162-167,共6页
Journal of Textile Research
基金
国家自然科学基金资助项目(51205294
61271008
51275363)