Text Region Detection in Images Using Connected Domain Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for locating text in images are inefficient, inaccurate, and lack general applicability due to complex implementation and high computational requirements, particularly in scenarios like video surveillance where text placement varies across devices.
Innovation Solution
A method involving binarization processing to set text pixels to specific colors, detecting connected domains, and extracting candidate boundaries based on text arrangement direction to accurately identify text regions without recognizing specific characters, thereby simplifying the process and reducing computational burden.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If character recognition technology (OCR) is used to detect text location, then text content can be recognized, but system complexity increases and processing efficiency decreases
Solution Approach 1:
The patent extracts only the text location information from the image without performing full character recognition. By using connected component analysis on binary images and extracting bounding boxes of text regions, the system obtains text location coordinates without needing to identify specific characters, thus reducing system complexity while maintaining the ability to locate text regions accurately
Solution Approach 2:
The patent creates a simplified binary copy of the original image through thresholding, where text regions are represented as connected components. This binary representation preserves the spatial location information of text while eliminating the need for complex character recognition, allowing the system to work with a simplified version of the image data
2Measurement precision
If mathematical models based on large data are used for text location detection, then detection capability is improved, but computation requirements become excessive
Solution Approach 1:
The patent uses simple, computationally inexpensive operations such as binary thresholding and connected component analysis instead of complex neural network models. These lightweight algorithms require minimal computational resources and can be executed efficiently on embedded devices, achieving adequate text location detection without the excessive computation requirements of large data models
Solution Approach 2:
The patent replaces complex computational models with simpler image processing operations. By substituting mathematical models based on large data with straightforward connected component analysis on binary images, the system achieves text location detection with significantly reduced computation requirements while maintaining practical accuracy
3Measurement precision
If traditional text detection methods are used, then text location can be identified, but general applicability across different hardware devices is poor
Solution Approach 1:
The patent implements a universal text detection approach that works across different hardware devices and text types. By using device-independent image processing operations (binary thresholding and connected component analysis) rather than device-specific parameters, the system achieves general applicability while maintaining text location identification capability across various scenarios including surveillance cameras and mobile devices
Data Source
AI summary
A method for locating text in an image is provided. Binarization processing is performed on the image, to set pixel points corresponding to text in the image to a specified color, and a binary image is obtained. A connected domain of pixels of the specified color in the binary image is detected, to obtain a candidate connected region. Connected domain candidate boundaries corresponding to a text region are extracted from the binary image, and the extracting is based on boundaries of the candidate connected region in the binary image and a text arrangement direction. The text region is obtained in the image by using the connected domain candidate boundaries.