Multiple Neural Networks for Confused Grapheme Classification
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep neural networks used for recognizing text in images require significant resources and processing time, making them inefficient and costly for accurate grapheme classification, especially when dealing with large alphabets or new languages.
Innovation Solution
Implementing multiple neural networks with fewer layers, each trained to recognize sets of commonly confused graphemes, reducing the complexity and resource requirements, and allowing for efficient text identification on conventional hardware without the need for specialized graphics processing engines.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deep neural network with a large number of layers is used to recognize all graphemes, then the recognition accuracy is improved, but the processing time and resource consumption increase significantly
Solution Approach 1:
The patent divides the grapheme recognition task into multiple segments by creating separate neural networks for different sets of confused graphemes. Each neural network is trained to recognize a specific subset of graphemes that are visually similar to each other, rather than using a single deep network to recognize all graphemes. This segmentation reduces the complexity and processing time of each individual network while maintaining overall recognition accuracy.
2Measurement precision
If a deep neural network with a large number of layers is used to recognize all graphemes, then the recognition accuracy is improved, but the hardware requirements and system cost increase
Solution Approach 1:
The patent segments the recognition system into multiple independent neural networks, each handling a specific subset of graphemes. This allows the system to achieve high recognition accuracy without requiring a single complex deep network, thereby reducing hardware requirements and making the system more accessible for deployment on conventional hardware.
3Device complexity
If a single neural network is used to recognize all graphemes, then the system is simpler to implement, but the processing efficiency decreases
Solution Approach 1:
The patent divides the grapheme set into multiple subsets based on visual similarity, with each subset handled by a dedicated neural network. This segmentation improves processing efficiency by reducing the search space and computational load for each network, while the overall system remains relatively simple to implement through modular architecture.
Solution Approach 2:
The patent performs preliminary classification to identify the set of confused graphemes that the input grapheme belongs to, before routing it to the appropriate specialized neural network. This preliminary action optimizes processing efficiency by ensuring that each grapheme is processed by the most suitable network, reducing unnecessary computations.
4Speed
If multiple specialized neural networks are used for different grapheme sets, then the processing speed is improved, but the system complexity increases
Solution Approach 1:
The patent segments the grapheme recognition task into multiple specialized neural networks, each optimized for a specific subset of graphemes. This segmentation improves processing speed by reducing the computational burden on each network, while the system architecture remains manageable through modular design and clear organization of grapheme sets.
Data Source
AI summary
A classification engine stores a plurality of neural networks in memory, where each neural network is trained to recognize a set of confused graphemes from one or more sets of confused graphemes identified in recognition data for a plurality of document images. The classification engine receives an input grapheme image associated with a document image comprising a plurality of graphemes, determines a set of recognition options for the input grapheme image, wherein the set of recognition options comprises a set of target characters that are similar to the input grapheme image, selects a first neural network from the plurality of neural networks, wherein the first neural network is trained to recognize a first set of confused graphemes, and wherein the first set of graphemes comprises at least a portion of the set of recognition options for the input grapheme image, and determines a grapheme class for the input grapheme image using the selected first neural network.