Hand-written font recognition method based on two-dimensional convolution dimension reduction

Abstract

The invention provides a hand-written font recognition method based on two-dimensional convolution dimension reduction, for solving the problem that in some complicated situations, such as high deformation, a current handwritten recognition algorithm is not well in high-dimensional data effect. The hand-written font recognition method based on two-dimensional convolution dimension reduction includes the steps: obtaining a special convolutional neural network by adding an optimization layer, so as to enable the network to be able to perform recognition and dimension reduction at the same time;designing a new linear determination analysis target function to reduce the complexity of the optimization process, and lowering the information dimension utilized by the network during the process ofrecognizing complex fonts and simplifying recognition by optimizing the function; and finally recognizing the hand-written font images by means of the trained network, to obtain the recognition result. As the hand-written font recognition method based on two-dimensional convolution dimension reduction uses the specially designed convolutional neural network to optimize the new linear determination analysis target function, and can obtain better recognition performance.

Description

technical field [0001] The invention belongs to the technical field of computer vision and graphics processing, and in particular relates to a handwritten font recognition method based on two-dimensional convolution dimensionality reduction. Background technique [0002] Handwritten font recognition is a widely concerned problem, and many solutions have been proposed, such as some early methods using template matching, such as in the literature "Chaturvedi S, Titre R N, Sondhiya N. Review of handwritten pattern recognition of digits and special characters Using feedforward neural network and Izhikevich neural model, Electronic Systems, Signal Processing and Computing Technologies (ICESC), 2014International Conference on.IEEE, 2014:425-428 "The method based on template matching mentioned in ", and the convolutional neural network now used for A series of algorithms for classification, such as et al. in the literature " D, Meier U.Multi-column deep neural networks for offl...

AI Technical Summary

Problems solved by technology

[0004] In order to overcome the deficiencies of the prior art, the present invention provides a handwritten font recognition method based on two-dimensional convolution dimensionality reduction to solve the problem that existing handwriting recognition algorithms are not effective for high-dimensional data under complex conditions such as high deformation , which mainly uses a specially designed convolutional neural network to optimize the linear discriminant analysis objective function, which can obtain better recognition performance

Images