CAJ | 학술논문

首先通过形态学边缘检测方法提取瓷砖花纹，然后采用分块提取特征的思想对瓷砖进行分类识别，即先将瓷砖分成4行4列16个子块，然后提取每子块两个稳定的Hu不变矩，共32个不变矩作为瓷砖分类的特征向量。对于分块后造成的矩的位置相关性问题，通过将每个训练样本分别旋转90°、180°、270°，然后再提取它们的矩特征向量的方法来解决。最后将提取到的特征向量输入BP神经网络进行分类识别。测试结果表明，本方法识别准确率高、运算速度快，能达到预期目的。
수선통과형태학변연검측방법제취자전화문，연후채용분괴제취특정적사상대자전진행분류식별，즉선장자전분성4행4렬16개자괴，연후제취매자괴량개은정적Hu불변구，공32개불변구작위자전분류적특정향량。대우분괴후조성적구적위치상관성문제，통과장매개훈련양본분별선전90°、180°、270°，연후재제취타문적구특정향량적방법래해결。최후장제취도적특정향량수입BP신경망락진행분류식별。측시결과표명，본방법식별준학솔고、운산속도쾌，능체도예기목적。
This paper involves morphological edge detection method by extracting tile pattern, and then use the block feature extraction for classification and recognition ideas on tile. Ceramic tile is firstly divided into 16 sub-blocks of 4 lines 4 and then extracted two invariant moments for each sub-block. So a total of 32 moments constitute the feature vector for tile classification. With the Hu moment is sensitive to location, because the division of the tile into sub-blocks, so it rotates each training sample 90,180 and 270 degrees in order to eliminate the in order to eliminate the negative influence. Finally, do the test with the extract feature vectors as the input of BP neural network. The results show that this method of identification is fast with high accuracy and achieve the desired purpose.