Lightness biased cartoon-and-texture decomposition for textile image segmentation

With the development of robust image processing tools in the textile industry, fabric designers are beginning to use feature extraction methods for both analysis and pattern design of fabric materials. In the design evaluation process, one of the basic problems is the efficient segmentation for textile fabric images. This is equivalent to partitioning the images into several meaningful regions that often correspond to units of design patterns, repeats, woven yarn or fibres. The main challenge in this problem is identifying robustly the boundaries of various components of fabric materials. In this paper, we propose a novel model to solve the problem. The model is established on the analysis of the characteristic of textile/fabric images. The main contributions of the model are: (1) a cartoon-and-texture decomposition process is incorporated into the model, which can reduce the influence of the random texture noise on the segmentation process; (2) to overcome the drawback of the lightness inconsistency for the segmentation process, a bias field function is introduced to measure the deviation degree between the cartoon image and the piecewise constant approximation of the cartoon image. Then, the regions of textile images can be more accurately estimated; (3) following the advantages of the fuzzy region competition based image segmentation models, we also use the fuzzy membership functions (FMFs) to indicate the regions of images. However, to restrain the FMFs from degeneration, a new penalty term on the FMFs is introduced in our model. In addition, by using the augmented Lagrange multiplier method and the Chambolle[U+05F3]s dual projection method, we derive an efficient algorithm to solve the model. Experimental results demonstrate that the proposed model can generate better segmentation results for textile images than classical FMF based models.