块离散余弦变换与灰色聚类相结合的表面瑕疵识别

Surface defect identification of block-discrete cosine transform combined grey clustering

针对工件或产品表面局部细微瑕疵不易识别、不利于在线检测问题,提出了一种块离散余弦变换(Block-Discrete Cosine Transform,BDCT)与灰色聚类相结合的表面瑕疵识别方法。该方法首先将图像分解成等尺寸的非重叠图像块,利用BDCT算法进行图像变换,提取待测图像的能量谱特性;然后将能量特性作为质量特征进行T2统计分析,初步查明待测图像表面瑕疵形状和性质;最后采用灰色聚类方法处理表面疑似瑕疵,进一步确定瑕疵具体位置和形状。为验证算法的有效性,设计了LED灯镜头表面瑕疵识别的检测方案和试验步骤,试验结果表明:本文所提出的算法相较经典的Otsu算法和分水岭算法在表面瑕疵正确识别率、速度等方面具有明显优势,可用于表面瑕疵的在线检测和识别。

It was difficult to identify and detect surface local slight defect of products online,a surface defect method was provided by Block-Discrete Cosine Transform( BDCT) combined grey clustering. An image was divided into no-overlapping image blocks and the energy spectrum was extracted by using BDCT algorithm. Then taken energy characteristics as quality feature to analyze T2 statistics,and found approximate surface defects form and features of image. Lastly suspected surface defects and shapes were further located by means of grey clustering. Testing scheme and procedures of LED lens surface defects were designed in order to verifying the validity of algorithm. Experimental results shows that presented algorithm has obvious advantages in speed and correct rate compared with classic Otsu and watershed algorithm,can be used to identify and detect surface defect of products online.