A SIMPLE NOVEL EFFECTIVE APPROACH TO DETECT CHARACTERS IN SCENE IMAGES - Egyptian Knowledge Bank

221216

A SIMPLE NOVEL EFFECTIVE APPROACH TO DETECT CHARACTERS IN SCENE IMAGES

Article

Last updated: 28 Dec 2024

Subjects

Abstract

In this paper, we adopt a new local binarization method as an important stage in text detection problem in scene image. Then we compare this new binarization method with our previous proposed binarization method which depends on using the Naïve Bayes classifier to classify the image pixels into foreground and background sets. After this, we apply the connected components analysis to get the morphological properties of each connected component (CC). These properties make used in the next stage to distinguish between text region and non-text region. Our proposed binarization method outperforms the well-known adaptive threshold method with respect to the Precession metric. That is in this paper, we target to satisfy two benefits; the first benefit is to enhance our previous model which be used to extract the characters' regions in scene images. This enhancement will be shown from the maximizing the Recall metric and this enhancement is fulfilled by considering a new binarization method, Bradley method. The second benefit to show the difference between our previous binarization method and the new one, Bradley method with respect to Recall and Precession metrics.

DOI

10.21608/aunj.2018.221216

Keywords

Binarization methods, Characters localization, Naïve Bayes classifier, Connected-components analysis

Volume

Article Issue

Related Issue

31544

Issue Date

2018-12-01

Receive Date

2022-02-22

Publish Date

2018-12-01

Page Start

Page End

Print ISSN

2812-5029

Online ISSN

2812-5037

Article File

AUNJ_Volume 47_Issue 2_Pages 58-73.pdf

PDF . 732.4kB

Link

https://aunj.journals.ekb.eg/article_221216.html

Detail API

https://aunj.journals.ekb.eg/service?article_code=221216

Order

221,216

Publication Type

Journal

Publication Title

Assiut University Journal of Multidisciplinary Scientific Research

Publication Link

https://aunj.journals.ekb.eg/

MainTitle