Text recognition method and device, electronic equipment and readable storage medium

Abstract

The invention discloses a text recognition method and device, electronic equipment and a readable storage medium. The method comprises the following steps: constructing a text recognition network model based on a convolutional neural network in advance; the text recognition network model comprises a feature extraction layer, a dense fusion block and a transcription layer. The dense fusion block comprises a first dense block, a second dense block and a convolution operation layer, the first dense block is connected with the second dense block, and the dense fusion block is used for performing fusion processing on the image features extracted by the feature extraction layer, the output features of the first dense block and the output features of the second dense block in different layers through connection operation; the transcription layer includes a classifier and a loss function layer. The accuracy and the recognition efficiency of text recognition can be effectively improved. And inputting the to-be-recognized text data into the trained text recognition network model to obtain a text recognition result, thereby effectively improving the accuracy and recognition efficiency of text recognition in the image data.

Description

technical field [0001] The present application relates to the technical field of computer vision, in particular to a text recognition method, device, electronic equipment and readable storage medium. Background technique [0002] Optical character recognition (OCR) is an important topic in the field of pattern recognition, and its purpose is to recognize text in images, which includes characters and numbers. Although OCR has been extensively studied for decades, accurately recognizing text in natural images is still a challenging task due to the complexity of background and content in images. In fact, characters and / or numbers may appear differently in different images due to variations in style, font, resolution or lighting. In recent years, with breakthroughs in the fields of computer vision and deep learning, end-to-end text recognition frameworks have been developed, which include complex two-step pipelines. The first step is to detect regions of text in the image, and...

AI Technical Summary

Problems solved by technology

However, since the pooling operation cannot be learned and may lose important information, the accuracy of the entire text recognition is not high

In addition, the traditional convolution operation is also computationally slow, so it should be replaced by a more efficient convolution operation

In addition, although dense blocks in DenseNet have good mobility and coupling of internal features, dense blocks and transition blocks in DenseNet are simply stacked together

In this way, the output features of each dense block are not well utilized, e.g., the hierarchical information of different layers is not fully utilized

Images