Fusion of color, texture and statistical features for enhancing content-based image retrieval

العناوين الأخرى

دمج سمات اللون، النسيج و السمات الإحصائية لتحسين عملية استرجاع الصور المعتمدة على المحتوى

مقدم أطروحة جامعية

al-Tarawinah, Ahmad Salim

مشرف أطروحة جامعية

al-Hasanat, Ahmad Bashir

أعضاء اللجنة

al-Kasasibah, Muhammad Sharari Zamil
al-Abbadi, Muhammad Ali Husayn
al-Uqayli, Salih Arashid

الجامعة

جامعة مؤتة

الكلية

كلية تكنولوجيا المعلومات

القسم الأكاديمي

قسم الحاسوب

دولة الجامعة

الأردن

الدرجة العلمية

ماجستير

تاريخ الدرجة العلمية

2015

الملخص الإنجليزي

Content-based image retrieval (CBIR) is one of the most debated topics in computer vision research, and has attracted a great deal of interest recently.

CBIR aims to retrieve similar images from an extensive unlabelled image database.

In this thesis we propose a method for CBIR systems that reduces the error rate and retrieves relevant images early in the process, with the ability to work on both color and grayscale images.

The proposed method scans an image using 8x8 overlapping blocks, extracting a set of the most discriminative statistical features from each block.

A histogram is created for each feature value, then these histograms are converted into probability density functions (PDFs) for each feature histogram.

Finally, these PDFs are fused together to obtain one discriminative features vector that represents the content of the image.

The most popular matching techniques are used in this thesis to compare the feature vectors.

Our experiments, conducted on several image databases, show the robustness of the proposed method, outperforming some of the most popular methods described in the literature.

In addition the proposed method was invariant to image rotation, and not affected much with image size.

The databases that were used in our experiments are Wang, Coil-100, IRMA-10000 and AT&T (faces) databases.

The results for precision are good on the Coil database, up to 0.998 at the first image retrieved, with an error rate of 0.002.

However, the results were less good on the Wang database, with 0.831 for precision and 0.169 error rate at the first image retrieved.

We also obtained good results on AT&T, up to 0.88 for precision and 0.12 error rate at the first image retrieved.

On IRMA-10000 the results were not good; IRMA- 10000 reflects the real environment of ultra sound images which often contain noise and cause many other problems.

التخصصات الرئيسية

تكنولوجيا المعلومات وعلم الحاسوب

عدد الصفحات

96

قائمة المحتويات

• APA
• MLA
• AMA

لغة النص

الإنجليزية

نوع البيانات

رسائل جامعية

رقم السجل

BIM-731645