OCR Processing Configuration for Runtime Hardware Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing optical character recognition methods require time-consuming and costly iterative processes to tailor processing operations for specific use-cases, particularly when characters have low contrast and visible contours, and are CPU-intensive, complicating hardware optimization.
Innovation Solution
A computer-implemented method that allows changing data processing configurations at runtime by loading them from a remote service, using precompiled functions optimized for specific hardware, enabling easier and faster iterations without replacing the entire software.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If processing operations are tailored for specific use-cases to achieve acceptable recognition performance, then recognition accuracy is improved, but the iterative process is time-consuming and costly
Solution Approach 1:
The patent pre-compiles multiple versions of processing operations during the software build process, each optimized for different hardware platforms and use-cases. These pre-compiled versions are stored in the software package and automatically selected at runtime based on the target device characteristics, eliminating the need for time-consuming iterative tailoring during deployment
Solution Approach 2:
The patent changes the parameter of compilation target by pre-compiling processing operations for multiple different hardware platforms (different CPU architectures, GPU types, mobile devices, desktop systems) during the build process. The appropriate pre-compiled version is then selected based on the runtime hardware detection, achieving optimal performance without iterative adjustments
2Productivity
If processing operations are optimized for particular hardware, then processing speed is improved, but device complexity increases
Solution Approach 1:
The patent creates a universal software package that contains multiple pre-compiled versions of processing operations, each optimized for different hardware platforms. The software automatically detects the target hardware and selects the appropriate pre-compiled version, achieving hardware-specific optimization without increasing the apparent complexity for the end user
Solution Approach 2:
The patent creates copies of the processing operations compiled for different target platforms. Instead of having a single complex optimization system, it uses multiple simplified copies of the same functionality, each pre-optimized for a specific hardware platform. The appropriate copy is selected at runtime based on hardware detection
Data Source
AI summary
A computer-implemented method for optical character recognition and corresponding data processing apparatus, computer program and computer-readable medium. The method includes the following: receiving image data of an input image having one or more characters and applying one or more data processing operations to the received image data, wherein at least one of the data processing operations is configured to return the one or more characters of the input image, wherein each data processing operation is selected from a predefined set of available data processing functions according to a processing configuration, wherein the available data processing functions are precompiled functions, and wherein the processing configuration can be changed at runtime.