Research On Image Generation Technology Based On Sketch

Image generation is a hot topic in the field of computer vision.As one of our country’s intangible cultural heritage.Thangka images have not yet achieved great development in the field of digital research.Carry out the method of drawing thangka images.This paper is based on neural network technology.We created a unique dataset.Based on triplet networks and generative adversarial networks,we propose two methods.Finally,the task of drawing a Thangka image with a simple sketch is realized.The main contributions of this paper are as follows.(1)This paper created a "Sketch-Edge-Thangka" dataset.We get Thangka images from the Internet.Obtain the edge map by the method of boundary extraction.Obtained the hand-free sketches by hand-drawing.Compared the three images corresponding to each other as triple data pairs.(2)This paper improved a sketch-based Thangka image retrieval network.The retrieval task is regarded as a fine-grained retrieval problem.The network model of this part is proposed by optimizing the triplet network and the attention network.This paper uses the cosine distance to evaluate the accuracy.The retrieval algorithm can provide reference and evaluation for the generated images.(3)This paper proposes a Thangka image generation network.Based on the idea of divide and conquer and generative adversarial network.This paper divides the generation of Thangka images into two parts: Thangka image skeleton generation and detail filling.Respectively optimizing different networks to improve the performance of each part,and finally improving the quality of the overall Thangka image generation.