Automatic identification of textile pattern consecutiveness based on similarity space

Abstract

This study presented a mechanism to automatically identify consecutiveness of textile patterns so as to help classify the ever-growing number of patterns. The pattern consecutiveness was first decomposed into two factors, including repeat angle and unit span. Then the pattern image was sliced into pieces on various angles with various spans, and the similarity degree of slices on each angle with each span was calculated to constitute the similarity space in which the peak values suggested the potential repeat angles and unit spans. The conjugacy of repeat angles was inspected in the similarity space, and the conjugate angles were advanced to distinguish the four-consecutives from the two-consecutives. Given the relativity of conjugate angles, the conjugate coefficient was brought forward to quantify the conjugate level of two perpendicular angles so as to find the optimal conjugate angles for the four-consecutives. The peak significance was proposed to discriminate sharp peaks against weak ones, and the similarity spaces of mono patterns were investigated to figure out the range of peak significance for the mono pattern, and thus the mono pattern could be identified with an absence of sharp peaks. A scheme for consecutiveness identification based on similarity space was finally carried out and implemented with computer programming, and it proved highly accurate and capable of tolerating certain flawed pattern images.