Algorithm Research Of Extended Chinese Character Sample Library Based On Style Transfer And Thin Plate Spline

With the rapid development of artificial intelligence,Chinese character recognition has played an important role in information entry and analysis,office automation and so on.Chinese character recognition technology has matured and achieved a lot of results.However,due to the complex light texture and different degrees of distortion and distortion of Chinese character samples collected in the natural environment,and the existing Chinese character sample inventory is small in sample size and single sample phenomenon,Therefore,the study of Chinese character recognition in the natural environment faces enormous challenges.The Chinese character recognition method adopted at this stage is based on deep learning,and the deep learning based method needs to provide a rich Chinese character sample library when performing network training.If the number of Chinese character sample libraries provided is small and the sample is single,the Chinese character training will have a problem of overfitting,and the trained network model is also difficult to be effectively applied in the actual scene.It can be seen that the establishment of a representative large-scale Chinese character database is the premise and basis for the study of Chinese character recognition in the natural environment.Since traditional methods(such as scanning,photographing,etc.)are used to collect Chinese character samples,there are enormous expenses in terms of manpower and material cost and time cost,and the efficiency of the calibration of Chinese character samples is also very low.Therefore,this paper proposes an algorithm based on fusion style migration and thin-plate spline expansion of Chinese character sample library,and automatically generates Chinese character samples by expanding the algorithm of Chinese character sample library.In this paper,the research status of Chinese character sample library amplification algorithm and Chinese character recognition technology at home and abroad is emphasized,and the method of amplifying Chinese character data samples is studied.Aiming at the influence of illumination and deformation on Chinese character recognition in natural environment,it is proposed to reduce the influence of illumination and deformation by amplifying the ray texture and deformation of Chinese characters.Therefore,this paper proposes an algorithm based on fusion style migration and thin-plate spline expansion of Chinese character sample library.The main work done in this paper is as follows:Aiming at the problem that the Chinese character sample library lacks the Chinese character texture sample image in the natural environment,the style migration is used to generate the Chinese character sample image with different illumination texture.Style migration is based on a convolutional neural network.First,the feature vectors of the content image(the original image of the Chinese character)and the style image(the light texture image)are extracted and their active blocks are calculated.Second,the calculation block is calculated for each content activation block.Matching style blocks;again,each content activation block is exchanged with its best matching style block;finally,the reconstructed complete content is activated to obtain stylized image results,thereby achieving the purpose of synthesizing the realistic texture of the Chinese character samples.For the problem that the Chinese character sample library lacks the Chinese character sample image with oblique and distorted deformation,the interpolation algorithm of thin plate spline deformation is used to generate the Chinese character sample image with tilt and distortion.Firstly,the SIFT feature matching is used in the original image and the target image to obtain the N pairs of feature points of the two images.Then,the N coordinate points of the original image are deformed to the corresponding positions in the target image by the thin plate spline deformation,and The entire space is used to achieve the deformation of the Chinese character image.Finally,the goal of generating Chinese characters with varying degrees of distortion and distortion is achieved.Aiming at the problem of insufficient Chinese character sample library and difficult collection in natural environment,this paper proposes an algorithm based on fusion style migration and thin plate spline deformation to generate richer and can simulate natural environment(light,texture,tilt,Distorted)A collection of Chinese character samples with real textures.Since the light shadow will change with the shooting angle when collecting Chinese character samples,the method of amplifying Chinese character samples proposed in this paper is to first synthesize the Chinese character texture image,and transform the Chinese character texture image to obtain both realistic texture and Samples of Chinese characters with varying degrees of deformation.Firstly,the block-based fast arbitrary style migration method is used to synthesize the Chinese character sample image with realistic texture.Secondly,the generated Chinese character sample image with realistic texture is generated by the interpolation algorithm based on thin plate spline deformation.With the distorted picture,the final Chinese character sample image has both a realistic texture and a tilt and distortion.The above algorithm is used to simulate the generation of Chinese characters in the natural environment,thereby creating a representative Chinese character data set.4.Using the camera to capture the Chinese character image in the natural environment,establish a small category of Chinese character sample library,and use the above algorithm to simulate and generate a Chinese character sample library of the same category;then train two newly established data sets through the deep learning network.The experimental results show that the algorithm based on style migration and thin-plate spline expansion of Chinese character sample database not only solves the problem of lack of Chinese character data set,but also improves the recognition rate of Chinese characters in the natural environment.