Iterative Recognition-Guided Thresholding for Complex Image Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional image capture and processing systems struggle to accurately extract information from digital images with complex backgrounds or overlapping foreground and background elements, particularly in documents with graphics or logos, leading to difficulties in distinguishing between foreground and background.
Innovation Solution
An iterative, recognition-guided thresholding process that identifies regions of interest and applies multiple thresholds to enhance contrast and extract features, using a learn-by-example classification technique to determine characteristic features and perform optical character recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If digital images are converted to bitonal or grayscale color depth prior to extraction, then processing efficiency is improved, but the ability to distinguish foreground from background deteriorates
Solution Approach 1:
The patent applies thresholding operations before conversion to bitonal/grayscale to pre-separate foreground and background elements. By performing segmentation in the color space first, the subsequent conversion to bitonal format preserves the discrimination accuracy because the separation has already been established.
Solution Approach 2:
The patent operates in multiple color spaces (RGB, HSV, LAB) and transforms between them to find the optimal dimension for foreground-background separation. By changing the color representation dimension, the system maintains discrimination accuracy while enabling efficient bitonal conversion for extraction.
2Productivity
If conventional thresholding is applied to images with complex backgrounds, then processing speed is maintained, but extraction accuracy deteriorates
Solution Approach 1:
The patent segments the image processing into distinct stages: color space conversion, thresholding in optimized color space, and final bitonal conversion. This segmentation allows each stage to be optimized independently, maintaining processing speed while improving extraction accuracy through targeted thresholding operations.
Solution Approach 2:
The patent dynamically adjusts thresholding parameters based on the specific characteristics of the input image, including background complexity and foreground properties. By changing parameters adaptively rather than using fixed thresholds, the system maintains high processing speed while achieving accurate extraction even from complex backgrounds.
3Measurement precision
If multiple thresholding operations are performed to enhance contrast, then extraction accuracy is improved, but processing time increases
Solution Approach 1:
The patent performs color space conversion and preliminary thresholding operations before the main extraction process. By preparing the image data in advance with optimal contrast enhancement, the subsequent extraction requires fewer iterative operations, reducing overall processing time while maintaining high accuracy.
Solution Approach 2:
The patent uses color space transformations (RGB to HSV/LAB) to enhance contrast in specific channels that are most discriminative for the given image. This dimensional transformation allows single or dual thresholding operations to achieve the same effect as multiple thresholding operations would in the original color space, reducing processing time.
Data Source
AI summary
According to one embodiment, a computer-implemented method comprises determining one or more features of an object class from a plurality of reference images, each reference image independently depicting at least one reference object belonging to the object class. The method further comprises determining, from among the one or more features of the object class, a set including at least one characteristic feature of the object class, where the characteristic feature(s) are each independently sufficient to identify the at least one reference object as belonging to the object class using a classification technique. The method further comprises determining one or more features of a test object from within one or more test images, each test image depicting the test object. The method comprises determining whether one or more of the features of the test object match or correspond to one or more of the characteristic features of the object class.