Simulation Of Embroidery Style Based On Convolutional Neural Network

Since the Non-photorealistic rendering proposed in the 1990s,it has been focusing on the simulation of different artistic styles and devoted to the performance of hand-painted styles and then became the research hotspot in computer graphics.Researchers have successfully simulated the oil painting,pencil painting,abstract painting,chalk painting and so on.And the results also have been widely used in advertising,online education,cultural industry and some other fields.With the deepening of research and the expanding of application area,the methods of simulating the Chinese cultural artworks have been proposed gradually.Including the ink and wash painting,calligraphy,pyrograph and so on.However,the simulation methods of the Chinese embroidery is rare.Chinese embroidery,the dazzling pearl in the history of Chinese civilization,it punctures on the cloth with needle and finally becomes into colorful patterns.Almost all the present methods for the simulation of Chinese embroidery are basedon mathematical modeling.They've built the database for each kind of needle,process the image step by step.But these methods can't make sense when they were used to the other kinds of embroidery.To remedy the deficiency of the current digital embroidery algorithm in arousing the original linear and stereo perception,this paper proposes an artificial system based on deep-learning and convolutional neural network to synthetize the style of embroidery.We process the content image and the embroidery style image by using two networks which were trained by different datasets respectively.This paper works with the following methods:First,we input the content image and the embroidery style image,perform image semantic segmentation which is based on conditional random field to separate the foreground and the background of both images and construct masks by image binaryzation.Then,we convert the color spaces of both input images from RGB into YIQ,extract the features of embroidery by VGG19 and transfer the content image into embroidery style in the foreground by using mask.Finally we blend the foreground image which has the embroidery style to the background image such as the real cloth and get the complete embroidery style image,meanwhile emphasizing the gorgeous colors and the stereoscopic textures of the embroidery.This study proposes a method which based on image semantic segmentation and style transfer to generate embroidery art style image.We make the original image into the style of embroidery and emphasize the gorgeous colors and the stereoscopic textures of embroidery through this algorithm.Experimental results have shown that the proposed method can generate images into embroidery style effectively and lays a foundation for digital inheritance of the traditional embroidery.