Automatic Image Annotation

Automatic Image Annotation and Retrieval Using the Joint

Composite Descriptor

Konstantinos Zagoris, Savvas A. Chatzichristofis, Nikos Papamarkos and Yiannis S. Boutalis

Department of Electrical & Computer EngineeringDemocritus University of Thrace, Xanthi, Greece

kzagoris@ee.duth.gr

Problem Definition Today, capable tools are needed in order to

successfully search and retrieve images. Content-based Image Retrieval Techniques

have been used such as img(Anaktisi) and img(Rummager) which employ low-level image features such as color, texture and shape in order to locate similar images.

Although the above approaches are successful, they lack the ability to include human perception in the query

Proposed Technique A new image annotation technique A keyword-based image retrieval system Employ the Joint Composite Descriptor Utilizes two set of keywords One Set consists of colors – keywords The other Set consists of words The queries can be more naturally

specified by the user

The Keyword Sets

Keyword Annotation Image Retrieval SystemThe Overall Structure of the Proposed System

Joint Composite Descriptor Belong to the family of Compact

Composite Descriptors More than one feature at the same time,

in a very compact representation Global Descriptor (global image low

level features) Fusion of the Color and Edge Directivity

Descriptor (CEDD) and the Fuzzy Color and Texture Histogram (FCTH)

CEDD and FCTH Descriptors The CEDD length is 54 bytes per image while

FCTH length is 72 bytes per image. The structure of these descriptors consists of

n texture areas. In particular, each texture area is separated into 24 sub regions, with each sub region describing a color.

CEDD and FCTH use the same color information, as it results from 2 fuzzy systems that map the colors of the image in a 24-color custom pallete.

CEDD and FCTH Descriptors

CEDD FCTH

Color Information given by the two descriptors comes from the same fuzzy system.

CEDD uses a fuzzy version of the five digital filters proposed by the MPEG-7 Edge Histogram Descriptor.

FCTH uses the high frequency bands of the Haar wavelet Transform in a fuzzy system.

So, the joining of the descriptors is accomplished by the fusion-combination of the texture areas carried by each descriptor.

CEDD and FCTH Descriptors

Each descriptor can be descripted in the following way:

Joint Composite Descriptor (JCD)

with

Joint Composite Descriptor (JCD)

Color Similarity Grade (CSG) Defines the amount of corresponding

color in the image It is calculated from the JCD for each

color keyword

Color Similarity Grade (CSG)

Portrait Similarity Grade (PSG) Defines the connection of the image

depiction with the corresponding word It is calculated from the normalization of

a trained Support Vector Machines (SVM) Decision Function

For each word a SVM is trained using as training samples the JCD values from a small subset of the available image database

Support Vector Machines (SVMs) Based on statistical learning theory Separate the space that the training samples are

resided in two classes The new sample is classified depending where in

the space residues

Portrait Similarity Grade (PSG) In this work, we used the following

equation to determine the membership value of the sample x to the class 1:

1 1100 max , 0

1 11 13 3

1 1100 1 max , 0

1 11 13 3

f x f x

f x f x

if f xe e

R x

if f xe e

Keyword Annotation Image Retrieval System User can employ a number of keywords from both sets

in order to describe the image. For each keyword , and initial classification position is

calculated based on Manhattan distance and its PSG or its CSG.

Then the final Rank is calculated for each image based on:

where N is the total number of images in the database For more keywords the sum of the Ranks for each

keyword is calculated:+

ImplementationThe proposed work is implemented as part of the img(Anaktisi) web application at http://www.anaktisi.net

Evaluation The Keyword based image retrieval is

tested in the Wang and the NISTER databases

Image Retrieval Results using the keyword “mountain”

Image retrieval results using the JCD of the first image as query.

Conclusions A automatic image annotation method which

maps the low level features to a high level features which humans employ.

This method provides two distinct sets of keywords: colors and words.

The proposed project has been implemented and evaluated on the Wang and NISTER databases.

The results demonstrate the method's effectiveness as it retrieved results more consistent with human perception.

Ευχαριστώ