Character Recognition In Shopping Platform Commodity Information Picture

With the rapid development of internet,more and more people are used to shopping on e-commerce platform.The simple,vivid and accurate visual information of commodity information picture brings convenience to consumers and challenges to the managers of e-commerce websites.Because the expression form of the characters in the pictures of commodity information is optical characters interfered by character diversity,background texture and other conditions,which can not be directly searched and processed by the computer,this causes some bad businesses to show the illegal commodity information to consumers in the form of pictures and avoid the network supervision.Therefore,it is of great practical value to study how to accurately extract text information from these e-commerce images.Automatic extraction of text information from commodity information pictures will help e-commerce enterprises to improve the recommendation efficiency of commodities,improve the level of after-sale guarantee and information-based supervision in the era of big data.This paper mainly focuses on the character location and recognition algorithm in e-commerce pictures.Based on a series of researches on image processing,character feature extraction and text location,the traditional feature extraction method of "edge detection and corrosion expansion" is abandoned.In this paper,we simulate the visual mechanism of the naked eye,and propose a method of character location through the steps of gray clustering,layer decomposition,denoising,etc.At the same time,combining with the neighbor search and statistical cutting technology,we cut the characters according to the geometric characteristics of Chinese characters.Finally,this paper uses the convolution neural network model,through the multi-layer convolution neural network,constructs a single character recognition model,successfully eliminates the background area of the picture,and realizes the extraction and recognition of text information in e-commerce pictures.In terms of model structure,this paper refers to a model of MNIST handwritten numeral recognition by convolution neural network.By increasing the number of convolution kernels and hidden nodes,changing the weight and other complex adjustments,the model structure is suitable for Chinese characters.In terms of activation function,ReLU function is selected,which greatly improves the model effect.In terms of preventing over fitting,the most commonly used dropoutmethod in deep learning network is used.When training the model,the final loss function of the model is adjusted,which greatly improves the prediction performance of the model.The results show that the method proposed in this paper has a good effect in practical application.In addition,a database of 35 fonts and 3062 characters including Song typeface,Kai typeface,Bold typeface,number and letter typeface has been established for the recognition of Chinese characters.A deep convolution neural network model is proposed for the non occluded printed Chinese characters.The model mainly includes two convolution layers,two pooling layers,one full connection layer,one hidden layer and one softmax.In this paper,by using a variety of innovative training methods,the recognition effect and generalization ability of the network are effectively improved.The recognition rate of the sample in the test data set reaches 97.5%.