Recognition Confidence Analysis of Handwritten Chinese Character with CNN

TL;DR

This study analyzes the confidence scores from CNN-based recognition of handwritten Chinese characters, demonstrating their usefulness in assessing reliability, identifying confusable pairs, and improving data quality.

Contribution

It introduces a method to analyze recognition confidence in Chinese character recognition and explores its applications in error detection and data correction.

Findings

CNN confidence scores correlate with recognition reliability

Confidence scores help identify confusable character pairs

Method aids in detecting mis-labelled or poorly written samples

Abstract

In this paper, we present an effective method to analyze the recognition confidence of handwritten Chinese character, based on the softmax regression score of a high performance convolutional neural networks (CNN). Through careful and thorough statistics of 827,685 testing samples that randomly selected from total 8836 different classes of Chinese characters, we find that the confidence measurement based on CNN is an useful metric to know how reliable the recognition results are. Furthermore, we find by experiments that the recognition confidence can be used to find out similar and confusable character-pairs, to check wrongly or cursively written samples, and even to discover and correct mis-labelled samples. Many interesting observations and statistics are given and analyzed in this study.