基于Fisher准则的分类器在皮革正反面分类中的应用

制鞋业中的皮革表面缺陷查找和排样主要靠手工完成，因此效率低下．基于图像处理技术的自动化缺陷查找及排样，可以较大地提高生产效率．皮革的缺陷查找和排样，主要在皮革的正面进行，因此皮革正反面的自动判别是后续处理的关键．因为皮革具有典型的纹理特征，纹理图像其特殊的像素空间分布方式有别于普通的灰度图像，因此常用的灰度图像分类特征对纹理图像的分类不适用．而基于灰度共生矩阵提取纹理图像的统计量组成的特征向量，并在Fisher判别准则的基础上设计一种线性分类器来对皮革纹理图像进行分类．实验结果证实，该分类器可有效地对皮革的正反面进行分类。

The Application of Classifier Based on Fisher Criterion on Classifying the Face and the Inverse of Leather

Inspecting defects of leather and arranging models in shoes industry are mainly performed manually,which have a poor efficiency. Inspecting defects and arranging models automatically based on image processing may improve the efficiency. In inspecting defects and arranging models performed on the face of leather, the classification of the face and the inverse of leather is important for later processing. Leather has typical texture and the texture images differ from common gray images in pixels space distribution. Therefore, general gray image characteristic for classifying are unusable for classifying texture images. This paper achieves characteristic of texture images based on co-occurrence matrix and designs a linear classifier based on Fisher criterion to classify leather images. The experimental results validated usefulness of the classifier introduced in this paper.