Color Texture Segmentation for Clothing Based on Finite Prolate Spheroidal Sequences

By means of a computer-aided fashion design system, a fashion designer can easily make changes to the various colors, textures and pattern prints of clothing design, and thus need not draw on paper a great number of drafts. One of the most difficult tasks in a computer-aided fashion design system is to separate the clothing of interest on a model from the background, so that changes can be applied to the material. There exist folds, shadows, diverse textures, etc. on the clothing which make the segmentation work difficult. In this paper, a color texture segmentation method for clothing segmentation is proposed. Color quantization is first performed to reduce the number of colors and shadow/highlight effects on the image. The color texture features are then extracted based on the finite prolate spheroidal sequences. By these features, a hierarchical coarse-to-fine segmentation method is used to separate the clothing from the backgrounds. Finally, post-processing is applied to obtain a smooth clothing boundary. Satisfactory experimental results have been achieved using the proposed approach.