2013 International Conference on Recent Trends in Information Technology (ICRTIT)
ISBN:978-1-4799-1024-3/13/$31.00 ©2013 IEEE 137
Analysis of Biologically Inspired Model for Object Recognition
S.Arivazhagan, R.Newlin Shebiah, P.Sophia, A.Nivetha
Department of Electronics and Communication Engineering,
Mepco Schlenk Engineering College, Sivakasi - 626 005
Abstract: Human visual system can categorize
objects rapidly and effortlessly despite the
complexity and objective ambiguities of natural
images. Despite the ease with which we see, visual
categorization is an extremely difficult task for
computers due to the variability of objects, such as
scale, rotation, illumination, position and
occlusion. This paper presents a biologically
inspired model which gives a promising solution to
object categorization in color space. Here, the
biologically inspired features were extracted by
log-polar Gabor Transform, aided by maximum
operation and convolution with Prototype patches
based on the saliency of the image. The extracted
features are classified by SVM classifier. The
framework has been applied to the image dataset
taken from the Amsterdam Library of Object
Images (ALOI) and the results are presented.
Keywords: Object Recognition, Log- Gabor
Transform, Biologically Inspired Model, SVM
I Introduction
Object recognition plays an important role in car
number plate recognition[1], face recognition for the
purpose of access control [2] and cancer recognition
[3], applications related to computer vision such as
video surveillance [4], image and video retrieval [5],
web content analysis [6], human computer
interactions [7] and biometrics[8] . In general, object
categorization is a difficult task in computer vision
because of the variability in illumination, scales,
rotation, deformation and clutter as well as the
complexity and variety of backgrounds.
W. Niblack et.al [9] proposed traditional appearance-
based approaches in the object recognition which
mainly use global low-level visual features such as
gray value, color, shape, and texture [9]. These
methods do not consider local discriminative
information and are sensitive to lighting conditions,
object poses, clutter, and occlusions.
J. Amores et.al. proposed Part-based models [10] that
make matches between particular patches and
interesting objects through various searching
schemes. In this framework, it is challenging to
robustly segment and find the meaningful parts, so
the spatial relationships of meaningful parts cannot
be duly modeled. G. Csurka, et.al introduced the
original bag-of-features based scheme [11] which is
efficient for recognition, but it ignores the spatial
relationship of features, and thus it is hard to
represent the geometric structure of the object class
or to distinguish between foreground and background
features. D. G. Lowe extracted Distinctive image
features from scale-invariant key-points. This is a
local feature based approach that combines the
interest point detectors and local descriptors with
spatial information. Representative local features
include scale-invariant feature transform (SIFT) [12].
Although these features are effective in describing
local discriminative information, they lack higher
level information, e.g., relations of local orientations.
T. Serre et. al. in [13] used a set of complex
biologically inspired features obtained by combining
the response of local edge-detectors that are slightly
position- and scale-tolerant over neighboring
positions and multiple orientations.
T. Serre et.al in [14] proposed a new set of scale and
position-tolerant feature detectors that are adaptive to
the training set. This approach demonstrates good
classification results on a challenging (street) scene
understanding application. Jim Mutch and David G.
Lowe in [15] builds on the approach of [14] by
incorporating some additional biologically-motivated
properties, including sparsification of features, lateral
inhibition, and feature localization. J. Mutch and D.
G. Lowe in [16] updates and extends the approach of
[15] by incorporating some additional biologically
motivated properties, specifically, sparsity and
localized intermediate-level features.
This paper is structured as follows: Section 2
describes about the proposed methodology. Results
and discussion is given in section 3. Last section
gives the concluding remarks.