快速CL-Isomap算法及其在印染检测中的应用

提出了一种基于聚类的带标志点的等距映射流形学习算法CL-Isomap,并可用于纺织印染品的自动化检测过程中.本算法针对印染图像的特点,先对图像集进行聚类,使得属于同一领域内的图像更加容易识别;然后使用标志点计算图像之间的距离,减少了计算量的同时,也更进一步分离了不同类的图像;最后使用MDS计算图像的低维嵌入.CL-Isomap算法对印染图像降维,使得图像数据量变小,检测速度提高,同时CL-Isomap方法在降维的同时并不改变图像原有的本质特征,使得检测的准确率得以提高.

Quick CL-Isomap algorithm and application in printing and dyeing detection

A CL-Isomap algorithm which is a fast clustering-based with landmarks algorithm is presented based on isometric mapping （Isomap） manifold learning algorithm. This algorithm used in textile printing and dyeing products for automated detection. For the characteristics of printing and dyeing images, this algorithm sets the image clustering firstly, making the images belong to the same area is more easily identifiable;then use the signs point to calculate the distance between the image for reducing the computation, and it further separat the different types of images. Finally, MDS is used in computing low-dimensional embedding images. At the same time, with the CL-Isomap algorithm to reduce the dimension does not change the essential characteristics of the original images, improving the detection accuracy.