UPM Institutional Repository

Zernike moments-local directional pattern fusion for content-based fish species image retrieval using momentgram and hue channel colour space


Citation

Osman, Noorul Shuhadah (2017) Zernike moments-local directional pattern fusion for content-based fish species image retrieval using momentgram and hue channel colour space. Masters thesis, Universiti Putra Malaysia.

Abstract

There is an increasing interest in the description and representation of fish species images. For that purpose, Content-based Image Retrieval (CBIR) is applied. Various techniques have been proposed for feature extraction to achieve good image representation and description result. One of them is the fusion of Zernike Moments (ZM) and Local Directional Pattern (LDP). ZM is rotation invariant and very powerful in extracting global shape feature, while LDP is texture and local shape feature extractor. However, existing works on ZM-LDP fusion are only used for gray-level images and are only invariant to rotation. While for fish images, colour plays an important role and the method should also be invariant to basic transformations such as rotation, translation, and scaling. This research proposes to improve the ZM-LDP method so that it will be able to extract colour features, be invariant to basic transformations, and further able to effectively represent the colour, shape and texture features for the fish-domain. The colour information property is obtained by computing the LDP on the Hue channel of the HSV colour space. The improved descriptor with colour information is tested on Fish4knowledge (natural image) image dataset consists of 27370 images and the proposed method has successfully achieved Mean Average Precision (MAP) of 77.62% and at the same time outperformed the other comparable methods. To achieve invariant to basic transformations, ZM-LDP fusion is improved by applying LDP on momentgram of the image. Retrieval experiment conducted on 27370 Fish4knowledge (mask image) image dataset have shown that the proposed method is able to achieve MAP of 91.3% and at the same time outperformed the other benchmark methods. These two proposed methods are then fused for content-based fish species image retrieval. Experiment is performed on 27370 Fish4knowledge (natural image) dataset, and the fused method has achieved MAP of 87.6%, which is higher than the benchmark methods. A statistical comparison based on the Two-tailed paired t-test has also been conducted and the proposed fused method has shown a significant improvement in retrieval performance.


Download File

[img] Text
FSKTM 2017 70 - ir.pdf

Download (1MB)

Additional Metadata

Item Type: Thesis (Masters)
Subject: Content-based image retrieval
Subject: Fishes - Detection
Call Number: FSKTM 2017 70
Chairman Supervisor: Mas Rina Mustaffa, PhD
Divisions: Faculty of Computer Science and Information Technology
Keywords: content-based image retrieval; shape; colour; feature extraction; fish; local feature; global feature; invariant to rotation, scaling and translation; local directional pattern; zernike moment; momentgram;
Depositing User: Ms. Nur Faseha Mohd Kadim
Date Deposited: 08 Sep 2020 07:24
Last Modified: 07 Jan 2022 07:31
URI: http://psasir.upm.edu.my/id/eprint/83246
Statistic Details: View Download Statistic

Actions (login required)

View Item View Item