Text recognition model training method and device, text recognition method and device and equipment
A text recognition and training method technology, applied in the field of text recognition, can solve problems such as the inability to meet the needs of the rapid development of computer vision tasks, the difficulty in achieving both recognition speed and recognition accuracy, and the inability of algorithms to be directly transferred to text detection.
- Summary
- Abstract
- Description
- Claims
- Application Information
AI Technical Summary
Problems solved by technology
Method used
Image
Examples
Embodiment 1
[0089] refer to figure 1 , the embodiment of the present invention provides a kind of training method of text recognition model, it is characterized in that, comprises:
[0090] Step S1: Construct the initial model, which consists of basic parts such as convolutional neural network (Convolutional Neural Networks, CNN), recurrent neural network (Recurrent Neural Network, RNN) and word embedding module. The convolutional neural network includes the first convolutional neural network Network and the second convolutional neural network, the initial model includes the first part and the second part, the first part and the second part are composed of a plurality of basic parts, wherein:
[0091] The first part is used to recognize the text content in the image, the first part has a first convolutional neural network and a recurrent neural network;
[0092] The second part is used to judge whether a given text is in a given image, and the second part has a word embedding module and ...
Embodiment approach
[0113] As an implementation manner, the second convolutional neural network includes four Block blocks arranged in sequence;
[0114] In this embodiment, by inputting the third text image into the second convolutional neural network, feature extraction is performed through each Block block in the second convolutional neural network, and after the feature map is obtained, the feature maps of the four Block blocks are respectively passed Downsampling or upsampling processing, respectively obtain the processed feature map, and the processed feature map of the four Block blocks has the same second size, in order to make the output of the second convolutional neural network meet the input of the word embedding module The same length, in this embodiment, the feature maps of the first two Block blocks in the four Block blocks are respectively down-sampled, and the feature map of the last Block block is up-sampled to obtain the processed feature map, so that The second size is 1 / 16 of...
Embodiment 2
[0125] refer to Figure 5 , the embodiment of the present invention provides a training device for a text recognition model, including:
[0126] an initial model building block for constructing an initial model;
[0127] The initial model training module is used to use the output of the first text image data through the cyclic neural network as the string input of the word embedding module, and based on the first text image data through the second convolutional neural network of the second part The first feature map obtained after is another input of the word embedding module, and the initial model is trained to obtain a convergent initial model;
[0128] A text recognition model acquisition module, configured to acquire a text recognition model based on the converged initial model;
[0129] Wherein, the initial model includes:
[0130] The first part is used to identify the text content in the image, the first part has a first convolutional neural network and a recurrent n...
PUM
Abstract
Description
Claims
Application Information
- R&D Engineer
- R&D Manager
- IP Professional
- Industry Leading Data Capabilities
- Powerful AI technology
- Patent DNA Extraction
Browse by: Latest US Patents, China's latest patents, Technical Efficacy Thesaurus, Application Domain, Technology Topic, Popular Technical Reports.
© 2024 PatSnap. All rights reserved.Legal|Privacy policy|Modern Slavery Act Transparency Statement|Sitemap|About US| Contact US: help@patsnap.com