Automatic extraction of flat sketch design element from clothing images using artificial intelligence

This study aims to develop a process that automatically extracts various flat sketch elements in vector format from clothing images. The approach of this study is to combine state-of-the-art image processing algorithms with the new algorithm devised in this study. First, a convolutional neural network based edge detection model trained with specially prepared fashion image data set was used to convert clothing images into edge maps. Then, a recurrent neural network based vectorization model and a rule-based image correction algorithm were used to convert the edge map into a vector image. Finally, a graph search algorithm was used to extract closed shapes from the vector image. As a result, the accuracy of edge extraction model has been improved by training with the special fashion image data set. The image correction rule was able to refine the vector images generated by the vectorization model. The graph algorithm was able to extract closed shapes from the vector image. This study is the first study to extract vector style flat sketch elements from clothing images using artificial intelligence and conventional computational geometry.