Automated flower species detection and recognition from digital images

Other Title(s)

الكشف و التمييز الآلي لأنواع الزهور من الصور الرقمية

Dissertant

al-Badarinah, Ala Abd Allah

Thesis advisor

Ahmad, Ashraf

Comitee Members

Muhanna, Muhanna
Kabnah, Khalid Abd al-Hafiz
al-Halabi, Yahya

University

Princess Sumaya University for Technology

Faculty

King Hussein Faculty for Computing Sciences

Department

Department of Computer Sciences

University Country

Jordan

Degree

Master

Degree Date

2016

English Abstract

Automated flower species recognition has been studied for many years.

Differences between these studies come from features which were extracted from the flower image, and the recognition algorithm that was used to recognize the flower species.

A new automated system was adapted to detect the flower region from the image and recognize its species.

By using a modified segmentation method, the features were extracted from the interest part only, so this increased the recognition accuracy.

The study aims at improving the flower recognition accuracy by extracting features from flowers’ leaf image.

Leaves are served as an important addition to the flower features, which is not been addressed by most of literatures.

Two datasets have been built, the first is Aalaa19 Dataset for flower only, and contains flowers from Jordan.

Including Iris Nigricans the national flower of Jordan, which is not addressed even in the famous Oxford17 dataset.

The second is FlowerAndLeaf21 Dataset for flowers with their leaves to study their effects on recognition accuracy.

Region growing segmentation was applied to extract flower and leaf region from their images.

Features based on color, texture, and shape have been extracted and used to train the Stochastic Gradient Descent (SGD) classifier with one versus all approach.

The results showed that the recognition accuracy for Aalaa19 Dataset is 92%.

The result of the second dataset FlowerAndLeaf21 showed that considering leaf in recognition increased the recognition accuracy from 71.1% to 77.31% with SGD classifier, while using random forest classifier increased the recognition accuracy from 83.5% to 85.5%.

Our proposed system outperforms several methods on Oxfoed17 Dataset; a class average recognition accuracy of 83.52% was achieved.

Main Subjects

Information Technology and Computer Science

No. of Pages

84

Table of Contents

• APA
• MLA
• AMA

Language

English

Data Type

Arab Theses

Record ID

BIM-693652