Classification of clothing with weighted SURF and local binary patterns

Abstract

This work proposes a recognition system for clothing classification by computer vision. The input is an image of the type of fashion catalog where the clothes are fully exposed with models showing their faces. For the preprocessing and features extraction the Bag of Features (BoF) is employed. There are four steps in the proposed classification method: (i) the cloth in an image is identified and located, then it is segmented by GrabCut; (ii) the area in the image of cloth is divided into three sub-windows (right side, middle, and left side); (iii) the feature extraction, Speed-Up Robust Features (SURF) and Local Binary Patterns (LBP) are applied to each sub-window to create a codebook; (iv) the classification is done by Support Vector Machine (SVM). Our dataset consists of total 1131 images out of which the training set is 991 images and the remainder is the testing set. We separate types into seven categories of clothing image which included, jacket, shirt, suit, sweater, t-shirt, polo-shirt and tank top. The result of the experiment illustrates that the proposed method can recognize types of clothing images accurately 73.57%.