Reaserch And Application Of Face Vectorization Algorithm Based On Sketch Extraction

Face line drawing is a face portrait composed of simple lines.In the fields of art creation,animation and web production,as well as criminal investigation,there is a wide demand for face line painting in vector format.Although there are abundant researches on face portrait and stylization,there is still a lack of work on transforming face photos into face line painting in vector format.In this paper,a face vectorization algorithm based on sketch extraction is studied,which aims to convert the input face image into a vector image in line drawing style.This paper focuses on maintaining the facial features and editable features in line drawing vector images and firstly proposes a two-stage face vectorization algorithm:In the first stage,a face line drawing is generated from the face image.a face line drawing extraction algorithm based on Generative Adversarial Networks is proposed.In order to solve the problem of face asymmetry on the left and right sides in face line painting,self-attention mechanism is introduced to maintain facial features and reduce the influence of background on the generated line drawings.In view of the fact that there are a large number of redundant short lines in face line painting,a long cycle module is designed,which uses false face to generate line drawing and guides the training of the original model to remove redundant short lines effectively.At last,the dual training strategy is designed to make full use of the unpaired data to improve the performance of the model.The FID index was improved by 1.5 compared with the benchmark algorithm on the CUHK-LineDraw,and the average number of short lines in the generated face line drawing decreased by 26%.The second stage is to vectorize the line drawing of human face and output the vector drawing in scalable vector graphics format.This paper proposes a layered vectorization algorithm for face,which divides the facial features in face line painting,organizes them in layers,and vectorizes them separately to preserve the semantic information of the lines,separate the face attributes,and meet the needs of strong vector image editing.The vectorization algorithm of face line drawing proposed in this paper for the first time improves the editable property of the output vector image.Compared with the classical vectorization algorithm,it has outstanding advantages in accurately maintaining the line shape and the facial feature of face line painting.Finally,this paper implements a face vectorization system based on WeChat applet.Users can complete the entire process from taking photos to drawing face line vector based on mobile terminal and mobile network.By means of image quality optimization,background removal and other strategies,the system achieves excellent line vectorization performance both in public dataset and the selfie dataset.