Preclassifying method and system for Chinese handwriting character recognition

Abstract

According to the invention, it is a pre-classification method used in a Chinese handwritten character recognition system. The handwritten Chinese character recognition system performs pre-classification and fine classification on the features of the handwritten Chinese characters to recognize the handwritten Chinese characters. The method includes: extracting the first low-dimensional Chinese character feature of handwritten Chinese characters, and generating a first candidate word group; extracting the high-dimensional second Chinese character feature of the handwritten Chinese character for fine classification; Dimensionality reduction of the second Chinese character feature of the low dimension, obtain the second low-dimensional Chinese character feature, and generate the second candidate word group; then obtain the final candidate word by the intersection of the first candidate word group and the second candidate word group Group. For the same handwritten Chinese character, the present invention performs pre-classification through two sub-pre-classifiers, and utilizes two kinds of Chinese character features to screen out two different candidate word groups respectively. Thereby avoiding the shortage brought by simply using a pre-classifier and a Chinese character feature to screen candidate word groups.

Description

technical field [0001] The invention relates to a recognition method and system for handwritten Chinese characters, in particular to a pre-classification method and device for handwritten Chinese character recognition. Background technique [0002] Handwriting recognition (CHWR) for portable handheld devices is very different from Chinese handwriting recognition for desktop computers due to the limited computing power and storage capacity of handheld devices. For a recognition system for all Chinese characters, a fine classifier based on high-dimensional features of Chinese characters is generally required. The computational complexity and memory capacity requirements of such classifiers are too high to be directly applicable to handheld devices. [0003] Figure 4a with Figure 4b The pipeline of two existing pre-classifiers is shown. Figure 4a In step 71, the structural features of Chinese characters are first input, and then elastic matching (dynamic programming method...

AI Technical Summary

Problems solved by technology

[0005] Existing pre-classifiers only use one feature to screen candidate word groups, and make a compromise between speed and accuracy. When the speed is high, the accuracy is low, and when the accuracy is high, the speed is slow. Its comprehensive performance is difficult to meet the needs of handheld devices.

Images