Document Image Inclination Calculation Using Similar Character Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for obtaining the inclination of a character string in document images are inaccurate due to variations in character shape and size, leading to differences between true and calculated inclinations, especially when characters are not aligned properly or when regression calculations are complex.
Innovation Solution
An image processing apparatus that extracts a character group with similar shapes and sizes, calculates the inclination using position information from this group, and optionally selects frequently occurring characters to minimize shape variation effects, thereby reducing calculation complexity and improving accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If regression calculation is performed using coordinates of all characters in a character string, then the inclination value can be obtained, but the calculation amount becomes large and the accuracy decreases due to character shape variations
Solution Approach 1:
The patent extracts only the necessary characters (first and last characters of each word) from the character string for inclination calculation, removing unnecessary characters that would increase calculation complexity and reduce accuracy due to shape variations. This extraction principle directly resolves the contradiction by selecting a minimal subset of characters that provides accurate inclination information without requiring complex regression calculations on all characters.
2Productivity
If only the first and last characters are used for inclination calculation, then the calculation amount is reduced, but the accuracy may decrease depending on the selected characters
Solution Approach 1:
The patent performs preliminary selection of the first and last characters of each word before inclination calculation. By pre-identifying these critical characters that best represent the character string's inclination, the system achieves both high calculation efficiency and accurate inclination values without needing to process all characters or perform complex regression analysis.
3Quantity of substance
If characters with varying shapes and sizes are included in the calculation, then more characters can be used, but the difference between true inclination and obtained inclination increases
Solution Approach 1:
The patent applies local quality by treating different characters differently based on their position and importance. Specifically, it focuses on the first and last characters of each word, which have the most significant contribution to representing the overall character string inclination. This selective approach ensures high accuracy while maintaining reasonable calculation efficiency, resolving the contradiction between using more characters and maintaining precision.
Data Source
AI summary
The object of this invention is to provide an image processing apparatus in which, in processing of a document image read by a document reading device, an inclination of a character string in the document image which is recognized in character recognition is obtained more accurately. The image processing apparatus includes a similar character extraction portion which extracts and outputs a character group comprised of characters having a shape and a size that are same with or similar to each other from among characters constituting a character string comprised of a character recognized in optical character recognition from a document image read by a document reading device; and an inclination calculation portion which calculates an inclination value of the character string based on position information of each character of the character group output from the similar character extraction portion.