Video-Based Wetting Detection For Blended Fabrics

Textile scientists are seeking for automated ways to understand the wicking phenomenon of blended fabrics from recorded videos at the pixel level. In response to such need, we design a video-based method for detecting pixels that will become wet and for estimating the timestamps of wetting events, which is the first step toward characterizing the wicking phenomenon. Since the wicking behaviors of the blended fabrics can be very different from one yarn to another within a small spatial region, simple frame-level thresholding with morphological preprocessing steps does not fit this application scenario. In this paper, we analyze for each pixel the color variation along the time for the wetting event detection. We develop an iterative merging algorithm rooted from the likelihood ratio test to obtain a coarselevel timestamp. The timestamp is then refined using a parametric curve fitted to a small neighborhood. Experimental results show that our automated method can achieve satisfactory wetting detection performance when the generated binary wetting-event video is compared with the raw wicking video.