Text Recognition via Frequency Feature Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing text recognition systems face challenges in accurately detecting and recognizing text in natural scene images due to complex backgrounds and environmental interference, leading to inefficiencies in recognition processes.
Innovation Solution
A text recognition method involving the acquisition of high-frequency and low-frequency feature maps, followed by an M-level convolution process through cascaded convolution modules, merging of feature maps, determination of probability and threshold maps, and calculation of a binarization map to identify and recognize text areas.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR systems are used for text detection in natural scene images, then the system structure is simple, but the recognition accuracy deteriorates due to complex backgrounds and environmental interference
Solution Approach 1:
The patent segments the feature processing into distinct frequency components (high-frequency and low-frequency feature maps) and processes them through separate convolution paths. This segmentation allows the system to handle different types of information (edges/details vs. overall structure) independently, improving text detection accuracy while managing complexity through modular processing
Solution Approach 2:
The patent transforms the feature representation by creating multi-scale feature maps through convolution operations with different kernel sizes and strides. This dimensional transformation in feature space enables the system to capture text patterns at multiple scales, resolving the contradiction between accuracy and complexity
2Measurement precision
If multi-scale feature processing is applied to improve text detection in complex backgrounds, then recognition accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs preliminary feature extraction and frequency decomposition at the beginning of the processing pipeline, creating high-frequency and low-frequency feature maps that are then reused across multiple convolution operations. This preliminary action avoids redundant computations and reduces overall processing time while maintaining multi-scale analysis capabilities
Solution Approach 2:
The patent merges the high-frequency and low-frequency feature maps through convolution operations to create a comprehensive feature representation. By combining these processed features, the system achieves accurate text detection without needing to process each scale independently from scratch, thus reducing processing time
Data Source
AI summary
The text recognition method includes: acquiring a first high-frequency feature map and a first low-frequency feature map of a target image; performing an M-level convolution process on the first high-frequency feature map and the first low-frequency feature map by M cascaded convolution modules to obtain M pairs of target high-frequency feature map and target low-frequency feature map of the target image, where M is a positive integer; merging the M pairs of target high-frequency feature map and target low-frequency feature map to obtain a target feature map of the target image; determining a probability map and a threshold map of the target image based on the target feature map, and calculating a binarization map of the target image based on the probability map and the threshold map; and determining a text area in the target image based on the binarization map, and recognizing text information in the text area.