Handwritten Keyword Spotting via Segmentation and Hybrid Models
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for recognizing handwritten characters in documents face challenges due to the uniqueness of each individual's handwriting style, leading to difficulties in accurately identifying and extracting keywords from noisy and varied handwritten document images.
Innovation Solution
A method and system that performs word segmentation, character recognition, and keyword spotting using a combination of Generalized Hidden Markov Models, Gaussian Mixture Models, Hidden Markov Models, and Support Vector Machines, along with pre-processing techniques like skeletal graph analysis and fuzzy inference systems, to identify and reject non-keywords in handwritten documents.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional OCR methods are used for handwritten character recognition, then machine-printed character recognition accuracy exceeds 99%, but handwritten character recognition fails due to the uniqueness of each individual's handwriting style
Solution Approach 1:
The patent segments handwritten text into individual character images, extracting features from each character separately. This segmentation allows the system to handle the variability of handwriting styles by analyzing individual character properties rather than requiring recognition of entire words or sentences, thereby improving adaptability while maintaining precision.
Solution Approach 2:
The patent transforms handwritten character images into numerical feature vectors by extracting parameters such as pixel density, aspect ratio, and structural characteristics. This parameter transformation converts variable handwriting styles into comparable numerical data, enabling the system to adapt to different writers while maintaining consistent recognition accuracy through mathematical comparison.
2Productivity
If keyword spotting is performed on noisy handwritten documents, then information extraction automation is achieved, but recognition accuracy deteriorates due to noise and variability in handwriting
Solution Approach 1:
The patent performs preliminary preprocessing of handwritten document images before keyword recognition, including noise reduction, binarization, and character segmentation. These preliminary actions clean and standardize the input data, removing noise and variability that would otherwise degrade recognition accuracy during the automated information extraction process.
Solution Approach 2:
The patent introduces an intermediary feature extraction layer between the noisy handwritten input and the keyword recognition system. This intermediary layer transforms variable handwriting into standardized feature vectors, mediating the transition from noisy images to reliable recognition data, thereby maintaining accuracy while enabling automation.
3Loss of information
If full transcription of handwritten documents is performed, then complete text extraction is achieved, but processing time and complexity increase significantly
Solution Approach 1:
The patent extracts only the essential keyword information from handwritten documents rather than performing complete transcription of all text. By selectively extracting keywords that meet specific criteria, the system achieves sufficient information extraction for many applications without the time cost of transcribing entire documents, thereby reducing processing time while maintaining useful information retrieval.
Data Source
AI summary
A method and system for the spotting of keywords in a handwritten document, the method comprising the steps of inputting an image of the handwritten document, performing word segmentation on the image to obtain segmented words, performing word matching, and outputting the spotted keywords. The word matching itself consisting in the substeps of performing character segmentation on the segmented words, performing character recognition on the segmented characters, performing distance computations on the recognized characters using a Generalized Hidden Markov Model with ergodic topology to identify words based on character models and performing nonkeyword rejection using a classifier based on a combination of Gaussian Mixture Models, Hidden Markov Models and Support Vector Machines.