Lightweight OCR Networks for Mobile Text Recognition
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing OCR technologies are resource-intensive and not optimized for efficient text recognition on mobile devices, limiting their application in scenarios requiring lightweight solutions.
Innovation Solution
Employing pre-trained lightweight text detection and recognition networks, optimized for mobile terminals, to perform text detection and recognition using methods like distillation and quantization to reduce network size and improve efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If existing OCR technologies are used, then text recognition capability is achieved, but resource consumption is high and hardware requirements are heavy
Solution Approach 1:
The patent transforms the OCR model parameters from high-precision floating point to low-precision integer format, reducing the model size and computational resource requirements while maintaining acceptable recognition accuracy. This parameter transformation enables deployment on mobile devices with limited resources.
Solution Approach 2:
The patent extracts and removes redundant or less important features from the OCR model through feature selection and network pruning techniques, keeping only the essential components needed for accurate text recognition. This extraction process reduces model complexity and resource consumption.
2Reliability
If existing OCR technologies are used, then text recognition capability is achieved, but device complexity and hardware capacity requirements increase
Solution Approach 1:
The patent changes the numerical precision parameters of the neural network from 32-bit floating point to 8-bit or 16-bit integers, significantly reducing the memory footprint and computational complexity while maintaining functional equivalence for text recognition tasks.
Solution Approach 2:
The patent creates a simplified copy of the original OCR model through knowledge distillation, where a large teacher model trains a smaller student model. The student model replicates the essential recognition capabilities with reduced complexity, enabling deployment on resource-constrained devices.
3Productivity
If model size is reduced through optimization, then efficiency on mobile devices improves, but recognition accuracy may deteriorate
Solution Approach 1:
The patent implements feedback mechanisms during the quantization process where the model performance is continuously monitored and the quantization parameters are adjusted iteratively to minimize accuracy loss while maintaining computational efficiency improvements.
Solution Approach 2:
The patent applies different quantization precision levels to different parts of the neural network based on their importance, with critical layers maintaining higher precision and less critical layers using lower precision, thereby optimizing the balance between accuracy and efficiency.
Data Source
AI summary
The present disclosure provides a method and apparatus for recognizing a text, a device and a storage medium, and relates to the field of deep learning technology. A specific implementation comprises: receiving a target image; performing a text detection on the target image using a pre-trained lightweight text detection network, to obtain a text detection box; and recognizing a text in the text detection box using a pre-trained lightweight text recognition network, to obtain a text recognition result.