Session T3B: OCR and Applications

Computer Vision — ACCV'98

Volume 1351 of the series Lecture Notes in Computer Science pp 160-167

Date:

Evaluation and application of recognition confidence in OCR

  • Xiaofan LinAffiliated withImage Processing Division, Department of Electronic Engineering, Tsinghua University
  • , Xiaoqing DingAffiliated withImage Processing Division, Department of Electronic Engineering, Tsinghua University
  • , Youbin ChenAffiliated withImage Processing Division, Department of Electronic Engineering, Tsinghua University
  • , Jinhui LiuAffiliated withImage Processing Division, Department of Electronic Engineering, Tsinghua University
  • , Youshou WuAffiliated withImage Processing Division, Department of Electronic Engineering, Tsinghua University

* Final gross prices may vary according to local VAT.

Get Access

Abstract

Recognition confidence plays an important role in the selection of rejection threshold and the combination of multiple classifiers. In this paper, we first present a systematic theory on classifier's confidence, which includes the definition, the concept of generalized confidence, optimal rejection theorem and the relationship between confidence value and recognition rate. Then we propose a method for the evaluation of recognition confidence. The theory and method are strongly supported by the practice in handwritten numeral recognition and off-line handwritten Chinese character recognition.

Key words

confidence optimal rejection handwritten numeral recognition off-line handwritten Chinese character recognition classifier combination