Document Image Recognition via Adaptive Block Thresholding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for recognizing document images from various types and shapes of information-printed matter are inaccurate, particularly for images with complex patterns, large-sized characters, or tables, leading to misjudgment.
Innovation Solution
A method and device that extract primary features from input images by determining unique feature points and applying classification criteria, including the total number of feature points and standard deviation of luminance values, to accurately classify images as document or non-document images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If edge detection method with threshold comparison is used to identify document images, then simple document images can be accurately detected, but complex pattern images are misjudged as document images
Solution Approach 1:
The patent divides the image into multiple blocks and performs edge detection on each block separately. By segmenting the image processing task, it enables more precise control over detection parameters for different regions, reducing misjudgment of complex patterns while maintaining accuracy for simple documents.
Solution Approach 2:
The patent dynamically adjusts the threshold parameter based on the standard deviation of luminance values in each block. Instead of using a fixed threshold, the threshold is adapted to local image characteristics, allowing accurate detection of document images while avoiding false positives from complex patterns with high edge density.
2Productivity
If fixed threshold edge detection is applied to all image areas, then processing is simple and fast, but accuracy decreases for images with varying character sizes and fonts
Solution Approach 1:
The patent applies different detection criteria to different blocks based on their local characteristics. Each block's threshold is determined by its own luminance standard deviation, creating locally adaptive detection parameters that maintain high accuracy across diverse document types without significantly increasing overall processing complexity.
3Device complexity
If the number of edge portions is used as the sole criterion for document image identification, then detection is straightforward, but images with large characters or tables are misclassified
Solution Approach 1:
The patent introduces luminance standard deviation as an additional parameter alongside edge count. This multi-parameter approach provides more discriminative power for classification, enabling reliable distinction between document images (including tables and large characters) and non-document images while maintaining reasonable detection complexity.
Data Source
AI summary
A method and a device for recognizing a document image, and a photographing method using the same are provided. The method includes extracting, from an input image, a primary feature, including information on feature points, by determining each pixel that is not similar to comparison pixels to be a feature point through a comparison of the pixel to the comparison pixels spaced a predetermined distance apart from the pixel, and classifying the input image as a document image or a non-document image by applying preset primary classification criteria, including criteria for the total number of the feature points to the primary feature.