Computer Vision for Supporting Fashion Creative Processes

Computer vision techniques are powerful tools to support and enhance creative workflows in fashion industries. In many cases, designer sketches and drawings, made with pen or pencil on raw paper, are the starting point of a fashion workflow. Then, such hand-drawn sketches must be imported into a software, to convert the prototype into a real-world product. This leads to a first important problem, namely, the automatic vectorization of sketches. Moreover, the various outcomes of all creative processes consist of a large number of images, which depict a plethora of products, from clothing to footwear. Recognizing product characteristics and classifying them properly is crucial in order to avoid duplicates and support marketing campaigns. Each feature could eventually require a different method, spacing from segmentation, image retrieval, to machine learning techniques, such as deep learning. Some state-of-the-art techniques and a novel proposal for line extraction and thinning, applied to fashion sketches, are described. Newly-developed methods are presented and their effectiveness in the recognition of features is discussed.