Gabor Filter Feature Vector Optimization for OCR Speed and Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing Optical Character Recognition (OCR) systems using Gabor filters face challenges with slow computation and low accuracy, especially for Asian characters due to large feature vectors or inadequate feature extraction, and require large learning sets for SVM systems, making them impractical or inaccurate.
Innovation Solution
A pattern recognition method that normalizes input images into matrices, generates feature vectors through sparse matrix multiplication using a Gabor function with specific parameters, and calculates probability densities to classify patterns efficiently, achieving a balance between feature vector size and redundancy for accurate and fast recognition.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a large feature vector is used to ensure accurate character identification, then recognition accuracy is improved, but computation speed deteriorates
Solution Approach 1:
The patent extracts only the most relevant features from the image by using Gabor filters to generate a feature vector that captures essential character information while discarding redundant data. This selective extraction maintains recognition accuracy while reducing the overall feature vector size to improve computation speed.
Solution Approach 2:
The patent optimizes Gabor filter parameters (frequency, orientation, scale) to generate a feature vector with optimal size for both accuracy and speed. By carefully selecting filter parameters and the number of filters applied, the system achieves a balance where the feature vector is large enough to capture character details but small enough for efficient processing.
2Productivity
If a small feature vector is used to speed up computation, then computation speed is improved, but recognition accuracy deteriorates
Solution Approach 1:
The patent extracts only the most relevant features from the image by using Gabor filters to generate a feature vector that captures essential character information while discarding redundant data. This selective extraction maintains recognition accuracy while reducing the overall feature vector size to improve computation speed.
Solution Approach 2:
The patent optimizes Gabor filter parameters (frequency, orientation, scale) to generate a feature vector with optimal size for both accuracy and speed. By carefully selecting filter parameters and the number of filters applied, the system achieves a balance where the feature vector is large enough to capture character details but small enough for efficient processing.
3Ease of manufacture
If known Gabor filters are used for feature extraction, then feature extraction is performed, but the filters do not work adequately with subsequent classification steps
Solution Approach 1:
The patent modifies Gabor filter parameters and the structure of the feature vector to ensure compatibility with classification algorithms. The feature vector is designed with specific properties that match the requirements of subsequent classification steps, improving the overall reliability of the recognition system.
Solution Approach 2:
The patent implements a system where the feature extraction process is optimized based on feedback from classification performance. The Gabor filter parameters and feature vector structure are adjusted to improve classification accuracy, creating a feedback loop that enhances overall system reliability.
4Loss of information
If average on regions of Gabor filters is used to extract character direction information, then directional information is extracted, but the method involves a lot of calculation making OCR too slow
Solution Approach 1:
The patent extracts directional information directly from the Gabor filter responses without performing additional averaging operations over regions. By utilizing the inherent directional selectivity of Gabor filters and selecting key features from their outputs, the system obtains character direction information while avoiding computationally intensive averaging steps.
Solution Approach 2:
The patent applies partial averaging or selective sampling of Gabor filter responses rather than computing averages over all regions. This partial action approach maintains sufficient directional information for accurate recognition while significantly reducing the computational burden compared to complete regional averaging.
Data Source
AI summary
A pattern recognition system and method which generates a feature vector by multiplying an image vector with a sparse matrix. The sparse matrix is generated from a Gabor function which is a sinusoidal wave multiplied by a Gaussian function. The Gabor function is a function of a set of parameters including a parameter related to the direction of the sinusoidal wave, a parameter related to a center of the Gabor function, and a parameter related to a wavelength of the sinusoidal wave. The wavelength takes at least two values, with a first wavelength value lower than or substantially equal to the distance between two adjacent centers of the Gabor function, and the first wavelength value is lower than a second wavelength value and higher than or substantially equal to half the second wavelength value.