Scene text recognition method based on multi-scale features

Abstract

The invention discloses a scene text recognition method based on multi-scale features, and the method comprises the steps: obtaining a scene text image data set, and carrying out the size adjustment of image data in the scene text image data set; constructing a text recognition model, and training the text recognition model by using the scene text image data set after size adjustment, wherein thetext recognition model comprises a multi-scale feature coding module and a multi-layer attention mechanism decoding module, the multi-scale feature coding module is used for acquiring character features of a plurality of scales, and the multi-layer attention mechanism decoding module combines a two-dimensional attention mechanism and a one-dimensional attention mechanism to fuse the character features of the plurality of scales to obtain a character recognition result in a scene text; and collecting a scene text image to be recognized, adjusting the scene text image to a uniform size, and inputting the scene text image into the trained text recognition model to complete recognition of the multi-scale characters in the scene text. According to the method, the multi-scale characters in the scene text can be quickly and accurately recognized.

Description

technical field [0001] The invention relates to the technical field of scene text recognition, in particular to a scene text recognition method based on multi-scale features. Background technique [0002] In recent years, with the rapid development of deep neural networks, the innovative application of artificial intelligence technology has been greatly promoted. Scene text recognition, as a part of artificial intelligence technology, has received extensive attention. Unlike optical character recognition in scanned documents, scene text recognition is very challenging due to various text fonts, low resolution, and susceptibility to light and shadow changes. The current mainstream scheme is summarized as using deep convolutional neural network to extract high-order features of images, using recurrent neural network to associate horizontal feature vectors, and finally using CTC (Connectionist Temporal Classification, connectionist temporal classification) or attention mechani...

AI Technical Summary

Problems solved by technology

Since the current scene text recognition technology only retains the output of the last layer of the deep convolutional neural network, its fixed and too large receptive field cannot adapt to the situation where characters of different sizes appear in the text at the same time, and in the case of generally small characters Under normal circumstances, more than half of its too large receptive field will fall in the background area, which seriously affects the performance of the recognizer.

Images