Word Spotting Without Character Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing word spotting methods require segmentation of characters, which can introduce errors and are computationally intensive, making them inefficient for identifying whole words in document images, especially in degraded images with broken or touching characters.
Innovation Solution
A method that characterizes word images by traversing them with a sliding window, extracting features based on the center of gravity, and using a 'Bag of Visual Words' approach without the need for character segmentation, allowing for classification of words without recognizing individual characters or using OCR.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If character segmentation is performed prior to feature extraction, then word recognition accuracy can be improved, but processing time and computational complexity increase significantly
Solution Approach 1:
The patent extracts only the necessary features (projection profiles, run lengths, stroke patterns) directly from the word image without performing full character segmentation. This selective feature extraction achieves sufficient recognition accuracy while avoiding the time-consuming segmentation process.
Solution Approach 2:
Instead of the conventional approach of segmenting characters first and then recognizing words, the patent inverts the process by directly analyzing word-level features from the complete word image, eliminating the need for character segmentation entirely.
2Adaptability or versatility
If OCR is applied to convert bitmap documents to character code format, then text searchability and classification capability are improved, but computational resources and processing time are excessively consumed
Solution Approach 1:
The patent extracts only the essential visual features needed for word identification (projection profiles, run lengths, stroke patterns) from the bitmap image, avoiding the computationally intensive full OCR process while still enabling text searchability and classification.
Solution Approach 2:
The patent performs partial OCR by extracting only the necessary features for word recognition rather than converting the entire image to character codes, achieving sufficient text processing capability with reduced computational overhead.
3Measurement precision
If character segmentation is performed in degraded images with broken or touching characters, then word identification can be attempted, but segmentation errors are introduced that reduce overall accuracy
Solution Approach 1:
Instead of attempting to segment broken or touching characters and then reconstruct words, the patent inverts the approach by directly analyzing word-level features from the complete word image, avoiding segmentation errors entirely while maintaining accuracy in degraded images.
Solution Approach 2:
The patent applies segmentation at the word level rather than character level, dividing the document into word images that can be recognized as complete units, which is more robust to character-level degradation and touching.
Data Source
AI summary
A method of characterizing a word image includes traversing the word image in steps with a window and at each of a plurality of the steps, identifying a window image. For each of the plurality of window images, a feature is extracted. The word image is characterized, based on the features extracted from the plurality of window images, wherein the features are considered as a loose collection with associated sequential information.