Abstract:

A method of image segmentation based on yarn color was proposed for the characteristics of backed weave images. Firstly, the image was transformed into a Lab color model, and hybrid median filter was used to remove scanning noise; then, a smoothing filter was improved by changing Gaussian weight with the allowable value of color difference and removing the shadows of overlapping structure and yarn textures with same color while maintaining the yarn color features; the gradient image based on color difference was extracted, and label image of fabric was obtained by watershed segmentation algorithm; finally, the index image of fabric with main color was obtained by merging the segmentation region with similar color. Experimental results showed that the algorithm can segment backed weave image accurately.

Key words: backed weave, image segmentation, color difference, smooth filtering, watershed algorithm