CAJ | 학술논문

由于织物图像纹理多样化及疵点类别较多，为了更有效地检测织物疵点，结合织物图像特性及借鉴人类视觉感知机理，提出一种基于纹理差异视觉显著性模型的织物疵点检测算法。该算法首先对图像进行分块，计算各个图像块LBP（ local binary pattern）纹理特征，与图像块平均纹理特征的相似度比较，进行显著度计算，从而有效突出了疵点区域。最后利用改进阈值分割算法，实现对疵点区域的定位。通过与已有视觉显著性模型进行比较，得出该算法更能有效地突出疵点区域；同时，分割结果与已有织物疵点检测算法相比发现，该算法具有更强的疵点检测及定位能力。
유우직물도상문리다양화급자점유별교다，위료경유효지검측직물자점，결합직물도상특성급차감인류시각감지궤리，제출일충기우문리차이시각현저성모형적직물자점검측산법。해산법수선대도상진행분괴，계산각개도상괴LBP（ local binary pattern）문리특정，여도상괴평균문리특정적상사도비교，진행현저도계산，종이유효돌출료자점구역。최후이용개진역치분할산법，실현대자점구역적정위。통과여이유시각현저성모형진행비교，득출해산법경능유효지돌출자점구역；동시，분할결과여이유직물자점검측산법상비발현，해산법구유경강적자점검측급정위능력。
In order to effectively detect defect for fabirc image with variety of defects and complex texture, a novel fab-ric defect detection scheme based on textural difference-based visual saliency model was proposed, which considered the characteristics of fabric image and human visual perception.First, the test image was split into image blocks, and tex-tural feature was extracted using LBP operator for each image block.Second, saliency was calculated by comparing their textural feature with the average texture feature.Finally, the threshold segmentation algorithm was used to localize the defect region.Comparing with the current saliency model, the proposed saliency model could effectively distinguish the defect.In addition, segmentation scheme was superior to the current defect detection algorithm in detection and lo-calization.