Performance evaluation of image retrieval algorithms using wavelet-based feature extraction:an experimental study

PERFORMANCE EVALUATION OF IMAGE RETRIEVAL ALGORITHMS

USING WAVELET-BASED FEATURE EXTRACTION:

AN EXPERIMENTAL STUDY

Hoang Nguyen-Duc1

, Thuong Le-Tien2

, Tuan Do-Hong2

, Cao Bui-Thu3

1Research & Development Department, Broadcast Research and Application Center – VTV

2Electrical-Electronics Engineering Department, Ho Chi Minh University of Technology

3Electronics-Telecommunication Division, Ho Chi Minh City University of Industry

Ho Chi Minh City, Vietnam

[email protected], [email protected], [email protected], [email protected]

ABSTRACT

In this article, we introduce an overview of performance

evaluation for the content-based image retrieval

algorithms (CBIR algorithms) and propose a set of

performance measures to evaluate retrieval effectiveness.

We select some CBIR algorithms using the wavelet-based

feature extraction which are introduced recently for

surveying. The texture image databases for benchmark are

used for experiments and compare to each other of

selected algorithms. The last part of this article,

conclusions for retrieval effectiveness evaluation are

given.

KEY WORDS

CBIR; content-based image retrieval; performance

evaluation; retrieval effectiveness;

1. Introduction

In content-based retrieval image, users describe the desired

content in terms of visual features (visual features of image

such as color or texture are represented using

multidimensional descriptors that may have tens or

hundreds of components), and the system retrieves images

that best match the description. Appropriate quantities that

describe the characteristics of interest must be defined,

algorithms to extract these quantities from images must be

devised, similarity measures to support the retrieval must

be selected, and indexing techniques to efficiently search

large collection of data must be adopted. Depending on the

design of CBIR algorithms that retrieval effectiveness will

be varied.

The performance evaluation of image retrieval

algorithms is a crucial problem in content-based image

retrieval. Many different methods for measuring the

performance of a system have been proposed by

researchers with the advantages and disadvantages were

analyzed by [6]. This paper will introduce some methods

are used to evaluate retrieval effectiveness of selected

CBIR algorithms for experimental surveying. These

methods are presented to be a sufficient survey about the

precision, recall, relevance, rank, and the best queries from

all queries in a class of database.

Recently, some CBIR algorithms using wavelet-based

feature extraction are introduced. We select nine CBIR

algorithms for experimental researches and performance

evaluation of them using two texture image databases.

These algorithms are designed using feature descriptors

based on: the Gabor wavelet transform [13, 2], the

contourlet transform [3, 8, 4], the non-subsampled

contourlet transform (NSCT) [10], the steerable pyramid

decomposition [9], the curvelet transform [12], the wavelet

transform [5], and the matching of many features using

wavelet-based features [7].

Through experiments, we will give some conclusions

about the application of the wavelet-based feature

extraction for image retrieval.

The remainder of this paper is structured as follow:

Next Section introduces more details about image

retrieval algorithms are selected for experiments, whilst

Section 3 represents and propose performance measures

are used for retrieval effectiveness and in Section 4, how

to calculate the performance measure is introduced, the

experimental results and the comparison between each

others from the algorithms are also given. Finally, Section

5 is devoted to concluding remarks.

2. CBIR Algorithms

2.1 The Typical CBIR Algorithm

The CBIR algorithm exploits features automatically

extracted from the image content that has the enormous

advantage of bypassing the need for keywords or other

annotation metadata explicitly associated to images.

Basically, a CBIR algorithm has an offline phase,

when the images will have their descriptions extract

features and will be stored in a database and may be

indexed. When extracting and indexing are ready, the user

can execute a query to the algorithms, in what is called the

online phase.

The entire process is illustrated on Figure 1.