Abstract
In this paper, an objective and reliable evaluation method of fabric pilling using a bottom-up visual attention model with higher accuracy is presented. Through analyzing visual attention mechanism and pilling characteristic, a new attention mechanism model is designed to detect pilling information, by which the contrast between pilling and background texture is enhanced in the saliency map constructed. And then, Otsu algorithm is adapted to segment the interesting region from saliency map, and area mean of foreground target is thought as the segmented threshold to separate pilling from interesting region. On this basis, extracting pilling features from the pilling binary images are classified through BP neural network. Experimental results show that compared with the traditional detection methods, the proposed method can evaluate objectively pilling grade effectively, and correct classification rate is over 94%.