Improved local binary pattern for real scene optical character recognition

Chu Sing Yang, Yung Hsuan Yang

Research output: Contribution to journalArticlepeer-review

37 Citations (Scopus)

Abstract

A strong edge descriptor is an important topic in a wide range of applications. Local binary pattern (LBP) techniques have been applied to numerous fields and are invariant with respect to luminance and rotation. However, the performance of LBP for optical character recognition is not as good as expected. In this study, we propose a robust edge descriptor called improved LBP (ILBP), which is designed for optical character recognition. ILBP overcomes the noise problems observed in the original LBP by searching over scale space, which is implemented using an integral image with a reduced number of features to achieve recognition speed. In experiments, we evaluated ILBP's performance on the ICDAR03, chars74K, IIIT5K, and Bib digital databases. The results show that ILBP is more robust to blur and noise than LBP.

Original languageEnglish
Pages (from-to)14-21
Number of pages8
JournalPattern Recognition Letters
Volume100
DOIs
Publication statusPublished - 2017 Dec 1

All Science Journal Classification (ASJC) codes

  • Software
  • Signal Processing
  • Computer Vision and Pattern Recognition
  • Artificial Intelligence

Fingerprint

Dive into the research topics of 'Improved local binary pattern for real scene optical character recognition'. Together they form a unique fingerprint.

Cite this