The Performance of Different Clustering Methods in the Objective Assessment of Fabric Pilling

In the present work, the performance clustering methods for classifying the pilling of knitted fabric and comparison among different classification algorithms is presented, Median- cut, K-mean, and Competitive Learning algorithms of K-mean were investigated in order to objectively evaluate fabric pilling. To achieve objective assessment, the fabric surface was scanned by a non-contact method using a laser triangulated device with no interfere to the fabric surface. The 2-D Fast Fourier Transform method considering low pass filtration and suitable cut-off frequency was applied to identify and separate pills from fabric fuzzy surfaces and fabric textures. Several fabric surface parameters were extracted but the number of pills, protruding geometrical volume, and area of pills were found to have better correlation with subjective evaluation. A pill grade intensity vector, comprising the three mentioned parameters was introduced for developing the clustering algorithms. The results show that K-mean and Competitive Learning algorithms of K-mean give a fair agreement between objective and subjective grades, while the Median-cut method has a poor correlation.