Handwritten Chinese Character Recognition Using Fixed-Point CNN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine learning approaches for recognizing handwritten Chinese characters are inefficient due to the need for large computer systems using floating point numbers, making it challenging to accurately read and understand variations in handwriting.
Innovation Solution
A CNN-based integrated circuit system that includes an input unit, memory, processing unit, and display unit, utilizing pre-trained filter coefficients for convolutional layers to extract features from input imagery data and classify handwritten Chinese characters, with a scalable and efficient architecture.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If floating point number based filter coefficients are used in prior machine learning approaches, then recognition accuracy can be achieved, but device complexity and computational resources required increase significantly
Solution Approach 1:
The patent transforms the filter coefficients from floating point numbers to fixed point numbers with specific precision requirements (e.g., 12-bit fixed point). This parameter transformation enables the use of simpler hardware architectures while maintaining sufficient computational precision for accurate handwritten Chinese character recognition, thereby reducing device complexity without sacrificing recognition accuracy.
2Measurement precision
If large computer systems are used for machine learning, then recognition capability is improved, but ease of operation and deployment difficulty increase
Solution Approach 1:
The patent divides the convolutional neural network into multiple stages with progressively smaller computational requirements. The architecture segments the processing into convolutional layers with decreasing filter sizes and depths, allowing the system to achieve high recognition capability while maintaining operational simplicity through a structured, manageable computational pipeline that can be deployed on more accessible hardware platforms.
3Measurement precision
If traditional machine learning methods are used, then recognition can be performed, but productivity and computational efficiency decrease
Solution Approach 1:
The patent replaces traditional software-based machine learning algorithms with a hardware-accelerated fixed point convolutional neural network implementation. This substitution of the computational approach enables significantly improved productivity and computational efficiency by leveraging optimized fixed point arithmetic operations and parallel processing capabilities, while maintaining the recognition accuracy required for accurate handwritten Chinese character recognition.
Data Source
AI summary
Apparatus for recognition of handwritten Chinese characters contains a bus, an input means connecting to the bus for receiving input imagery data created from a handwritten Chinese character, a Cellular Neural Networks or Cellular Nonlinear Networks (CNN) based integrated circuit operatively connecting to the bus for extracting features out of the input imagery data using pre-trained filter coefficients of a plurality of order convolutional layers stored therein, a memory connecting the bus, the memory being configured for storing weight coefficients of fully-connected (FC) layers, a processing unit connecting to the bus for performing computations of FC layers to classify the extracted features from the CNN based integrated circuit to a particular Chinese character in a predefined Chinese character set, and a display unit connecting to the bus for displaying the particular Chinese character. Greater than 95% recognition accuracy is achieved using multiple bi-valued 3×3 filter kernels as pre-trained filter coefficients.