10-04-2017, 09:34 PM
Grayscale Image Retrieval using DCT on Row mean,
Column mean and Combination
[attachment=303]
Abstract
Today in the age of information explosion, how to search appropriate data from huge information
pool has become vital issue. Images have giant share in this information pool. Because of easy
availability of imaging devices, millions of images are being added to image pool every day. Image
retrieval deals with searching relevant images from large image database. The paper presents novel
image retrieval techniques based on discrete cosine transform applied on row mean, column mean
and combination for feature extraction.
Introduction
In the past few years, there has been
tremendous growth in the database technology to
store, retrieve and process large number of
images [1],[15],[16]. Initially, images were
represented just by textual description and
observations. But these methods hardly captured
the vivid details of an image. This led to research
on the lines of automatically extracting features
of images for the purpose of efficient retrieval
and sequencing of images which is referred as
Content Based Image Retrieval (CBIR)
[3],[4],[14]. Main intention of CBIR is efficient
retrieval of images form a huge image database
based on some automatically extracted features.
These features are extracted from properties such
as shape, color and texture of query image and
the various images in the repository [7].
Discrete Cosine Transform
The discrete cosine transform (DCT) [5],[15]
is closely related to the discrete Fourier
transform. It is a separable linear transformation;
that is, the two-dimensional transform is
equivalent to a one-dimensional DCT performed
along a single dimension followed by a onedimensional
DCT in the other dimension.
DCT Row Mean Image Retrieval
Here first the row mean of query image is
obtained. Then the DCT row mean feature
vector of query image is obtained by applying
DCT on row mean. For image retrieval using
DCT row mean, these query image features are
compared with DCT row mean features of
image database by finding Euclidian distances
using the formula given as equation 1.
These Euclidian distances are sorted in
ascending order and result images are grouped
together to get the precision and recall using the
formulae as given below in equation 7 and
equation 8.
Results and Discussion
Five images from each category are provided
as a query and are compared to images sorted
according to values of Euclidian Distance with
cumulative increment of 2 images up to 100
images, to obtain precision and recall values. To
determine which technique of generating feature
vectors is best, the average values of precision
and recall of all the techniques are plotted as
shown in Fig. 3 and Fig. 4.