OCR Pre-localization via Pyramid Filter Tiling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current optical character recognition (OCR) systems face challenges in accurately identifying regions of interest (ROI) within documents due to variations in contrast, module values, and gradient angles, which affect the efficiency of character recognition.
Innovation Solution
The proposed method involves applying a pyramid filter to an image to create a filtered image, tiling it into defined-sized tiles, determining contrast, module values, and gradient angles for each tile, and identifying tiles with high contrast and module values to generate regions of interest (ROI) for further processing by an OCR decoding module.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If the entire image is processed for OCR without pre-localization, then the OCR system can process the complete document, but the processing efficiency decreases and irrelevant information increases the computational load
Solution Approach 1:
The image is divided into multiple tiles of defined size (e.g., 8x8 pixels or 16x16 pixels), and each tile is independently analyzed for contrast and module values. This segmentation allows the system to process only relevant portions of the image that contain characters, rather than the entire image, thereby improving processing efficiency and reducing computational load on irrelevant areas.
2Measurement precision
If pre-localization is applied to identify regions of interest, then the accuracy of character recognition improves, but the device complexity increases due to additional processing steps
Solution Approach 1:
The system performs preliminary analysis by applying a pyramid filter to the original image to create a filtered image, then divides it into tiles and calculates contrast and module values for each tile before actual OCR processing. This preliminary action identifies and isolates regions of interest that contain characters, allowing the subsequent OCR process to focus only on relevant areas, thereby improving accuracy while the added complexity is managed through efficient algorithm design.
3Reliability
If threshold criteria are used to filter tiles, then the relevance of selected regions improves, but the loss of information may occur if valid characters are excluded
Solution Approach 1:
The system uses adjustable threshold values for contrast and module values to filter tiles. By optimizing these parameters, the system can reliably identify regions with characters while minimizing the exclusion of valid characters. The pyramid filter and tile-based approach allow for flexible parameter tuning that adapts to different image characteristics, ensuring reliable ROI selection without excessive information loss.
Data Source
AI summary
A method of operation of an optical character recognition pre-localization system, comprising receiving an image having an original frame to be pre-processed for optical character recognition, applying a pyramid filter to the original frame in order to create a filtered image containing blobs to be analyzed as potential regions of interest that may contain characters, tiling the filtered image into a set of tiles of a defined tile size, and for each tile, determining a contrast, a module value and a gradient angle, identifying a first tile of the set of tiles of the filtered image, wherein the contrast and the module values of the first tile are higher than a respective contrast threshold and module threshold, and generating at least one region of interest of the original frame to be passed to an optical character recognition decoding module based on the first tile, and related systems.