Image Reading Device Inclination Correction Using Segmented Resolution Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image reading apparatuses face challenges in maintaining image quality during inclination correction, which degrades character recognition accuracy, especially when using reduced resolution images for correction, and require longer processing times and increased data amounts.
Innovation Solution
The apparatus employs a mode-setting unit to choose between quality-first and speed-first modes based on image resolution and character size, correcting inclination after resolution conversion to minimize quality degradation and reduce processing time.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If inclination correction is performed on a full-resolution source document image, then image quality is maintained, but processing time increases and data amount increases
Solution Approach 1:
The patent segments the processing into two distinct stages: first performing inclination detection and correction on a down-sampled low-resolution version of the image, then applying the same correction parameters to the full-resolution image. This segmentation allows rapid processing in the detection phase while preserving quality in the final output phase.
Solution Approach 2:
The patent creates a down-sampled copy of the source document image for inclination detection purposes. This copy serves as a surrogate that enables fast processing and parameter determination without compromising the quality of the final corrected image, as the correction parameters derived from the copy are applied to the original full-resolution image.
2Loss of time
If inclination correction is performed on a down-sampled image, then processing time is reduced, but image quality degrades
Solution Approach 1:
The patent separates the inclination correction task into two stages: parameter estimation on down-sampled images and application to full-resolution images. This ensures that the quality-degrading down-sampling operation is only used for parameter estimation, while the final correction operates on the complete quality image.
Solution Approach 2:
The patent performs preliminary inclination detection and parameter estimation on down-sampled images before applying the correction to the full-resolution image. This preliminary action on reduced data enables fast parameter determination that can then be applied to maintain quality in the final step.
3Productivity
If OCR is performed after inclination correction on reduced resolution image, then processing speed increases, but character recognition accuracy decreases
Solution Approach 1:
The patent segments the processing pipeline so that inclination correction is performed on down-sampled images while OCR is performed on the original full-resolution image. This segmentation ensures that the speed benefit of reduced resolution is captured in the correction phase, while the accuracy requirement for OCR is satisfied by using the complete quality image.
Solution Approach 2:
The patent performs preliminary inclination correction based on parameters derived from down-sampled images, then applies this correction to the full-resolution image before OCR. This preliminary action on reduced data enables fast parameter determination without compromising the quality of the image subjected to OCR.
Data Source
Figure 1
Figure 2
Figure 3
AI summary
An image reading unit (5) reads a source document image in 600 dpi, and a resolution conversion unit (84) converts the resolution of the image to 75 dpi. An inclination detection unit (82) detects inclination of the source document image. When a character conversion unit (85) is to execute character recognition after cropping, a mode setting unit (101) sets a quality-first mode. The inclination correction unit (82) corrects the inclination of the image of 600 dpi, according to the inclination of the source document image detected by the inclination detection unit (82). A document image clipping unit (81) clips out the source document image, and the resolution conversion unit (84) converts the resolution of the source document image to 200 dpi. Thereafter, the character conversion unit (85) executes the character recognition.