Research On Pencil Drawing Algorithm Based On Neural Network

With the development of society,science and technology,photography has become a way of recording lives,and people have began to pursue a variety kinds of photo styles.Since the 1990 s,non-photorealistic rendering technology has come into people's sight.Pencil drawing technology is one of the non-photorealistic rendering,which can draw out the artistic style of pencil drawing.Traditional methods are mainly based on the simulation of direction,thickness,length of strokes and the simulation of texture,but the result is rigid and single.In recent years,Convolutional Neural Network(CNN)and Generating Adversarial Network(GAN)have made rapid development and application in image processing.Therefore,this dissertation proposes a pencil drawing algorithm based on convolutional neural network.The main research contents are as follows:1.Research the development of non-photorealistic rendering technology,especially pencil drawing technology,including theories,advantages and disadvantages of traditional pencil drawing algorithm,and propose an improved algorithm.Reaearch the development of CNN and GAN,and their applications in image processing.2.A CNN-based pencil drawing generation method was presented.Fractional differentiation was employed to obtain original image contour information,CNN to obtain pencil drawing styles of different artists,the total loss function of the network was designed,and L-BFGS algorithm was used to synthesize pencil drawing image.The results show that this method can retain more details of the original image,and solve the problem that the traditionalpencil drawing algorithms have only one artistic style.3.A Generating Adversarial Network model was presented to solve the problem of lacking artistic styles in traditional pencil drawing algorithm results.Two groups of generation adversarial networks were designed.The first generating network was used to generate pencil drawing style images,adversarial network to learn the features of artists' real pencil drawing by using convolution neural network.The second generation adversarial network used the principle of cyclic consistency to retain the structure information of the original image.The distance deviation was optimized and the loss of detail control was added.The experimental results shown that this method can generate more artistic pencil drawing image and retain the structure information of the original image better.