Research On Synthetic Framework Of Geometric Free Stroke Based On Neural Network

Freehand sketches are an important medium for expressing and communicating ideas.However,creating a meaningful and understandable sketch drawing are not always an easy task,especially for users who have not been trained in painting.In order to help users without professional drawing skills to create the right sketches,more and more researches has begun to focus on the work of sketch generation,but most of these work is limited to the conversion of two-dimensional images,and there is no effective way to generate sketches of a 3D object.Existing methods for rendering 3D shapes into line drawings,such as Suggestive Contours,only consider the geometry-dependent and view-dependent information,thus lead to over-regulars or over-perfect results,which doesn't look like a human freehand drawing.Therefore,based on the above problems,this paper propose a method that can automatically generate freehand line drawing sketches from 3D objects under a given viewpoint: DeepShapeSketch.The core solution to this method is to use a deep recurrent neural network,which learns the functional mapping from the suggestive contours of a 3D shape to the more abstract sketch representation.This method drop the encoder of generator,i.e.,use only a decoder to achieve better stability of the sketch structure;and add a discriminator to distinguish between the generated sample and the real sample to facilitate the correctness of the sketch results generated by the generator.In addition,our framework provides a parameter ? for users to adjust the level of the freehand drawing style of sketches.We conduct a series of freehand sketch experiments,the results show that the sketches generated by the DeepShapeSketch method can correctly represent the input shape,and also has the freehand style created by real humans.To test the scalability of the DeepShapeSketch method for the particular application,several different hybrid datasets were used to retrain the neural network.The results show that these mixed datasets do not produce the expected results.Finally,we compare the DeepShapeSketch method with the existing SketchRNN method,which proves that DeepShapeSketch method can generate sketches that are more in line with human painting styles and can guarantee that the generated sketches have a similar viewpoint direction as the input.In addition,in order to verify the quality and style of the sketches generated by the existing methods,the paper conduct a user survey,and the user survey results also prove that the method can produce better quality sketch results.