Feature Extraction and Image Matching of 3D Lung Cancer Cell Image
Hizmawati Madzin
Multimedia Department
University Putra Malaysia
Serdang, Malaysia
hizma@fsktm.upm.edu.my
Roziati Zainuddin
Artificial Intelligence Department
University Malaya
Kuala Lumpur, Malaysia
roziati@um.edu.my
Abstract— The demand for automation in medical analysis is
continuously growing with large number of application in
biotechnology and medical research. Feature extraction and
image matching are important steps in analyzing medical cells.
In this research paper, we are concentrating on extracting and
matching features from a full 3D volume data of lung cancer
cell that was recorded with a confocal laser scanning
microscopy (LSM) at a voxel size of about (0.3μm)
3
. In order to
apply feature extraction on 3D cell image, the image is slices
into ten different viewpoints of 2D images with thickness of
each slice are about 0.1μm. An experiment has been done
based on local invariant features methods which are
HarrisLaplace method to extract features of each slices and
SIFT matching method to find and match same features in
each slices. The experiment shows that these methods can
extract the same features although in different viewpoints. This
research paper application can be served as preliminary step
for further research study in analyzing 3D structure of cancer
cell image.
Local invariant features; feature extraction; image
matching; 3D cancer cell image.
I. INTRODUCTION
In biological and medical research, the application of fast
volumetric imaging technique and aim of extracting
quantitative data from these images is obvious. Usual 3D
structure representations lack of concise information about
the object. As the representation of 3D objects is not
canonical and object occur at different spatial position and
viewpoint [1]. The demand for efficient image analysis
technique of 3D object is continuously growing. As for 3D
microscopy images, image analysis system is an essential yet
very challenging problem where the quantitative analysis of
morphology and cell phenotypes is needed. The difficult
part of microscopy image mostly come from the variability
and complexity of the images that caused by the intrinsic
properties of the samples as well as the staining and imaging
procedures [2]. Moreover microscopy image can be
complicated due to the imperfect staining or intrinsic
intracellular characteristics. The major objectives of image
analysis in biomedical engineering are to gather information,
screening or investigating, to diagnose, therapy and control,
monitoring and evaluation [3].
Feature extraction and image matching are important
steps that relate to each other in order to apply 3D medical
image analysis successfully. Local invariant features have
been found as well suited to matching, recognition and other
applications as they are robust to occlusion and content
changes [4]. According to Patrick [5], feature based
matching means finding the same features in different
images that represent in the same object. For this research
paper, Harris-Laplace detector method is used to extract
features from multiple viewpoints of confocal images and
SIFT method is used to identify and calculate the same
features found in each images.
3D volume data of lung cancer cell has been used in the
experiment of feature extraction and image matching. The
lung cancer cell image was taken from confocal laser scan
microscopy. In order to apply feature extraction, the 3D
cancer cell image is slices into ten slides of 2D images with
different viewpoints.
It is important to emphasize that the purpose this research
paper is to experiment whether local invariant features
methods is suitable to apply on multiple viewpoints of
confocal images. Therefore this research can be served as
preliminary step for further research study in analyzing 3D
structure on confocal microscopy image.
This paper is organized in four sections. The first section
is introduction, followed by the description of local
invariant features in section two. The next section will be
the methodology description of HarrisLaplace and SIFT
methods. The following section is the result of both
methods. The final section will be conclusion.
II. LOCAL INVARIANT FEATURES
Methods based on local invariant features have shown
promise good result for image analysis task. A local feature
is an image pattern which has different value or
characteristic from its instantaneous neighborhood and
associated with a change of image properties (intensity, color
and texture) simultaneously [6]. Invariant feature is a value
stay unchanged when a transformation (object’s position
and/or orientation changes) is applied [7]. Therefore, local
invariant feature is a new image representation that allows to
describe the objects/parts in any different transformation
(translation, rotation and scale). The features can be in
2009 International Conference of Soft Computing and Pattern Recognition
978-0-7695-3879-2/09 $26.00 © 2009 IEEE
DOI 10.1109/SoCPaR.2009.103
517
2009 International Conference of Soft Computing and Pattern Recognition
978-0-7695-3879-2/09 $26.00 © 2009 IEEE
DOI 10.1109/SoCPaR.2009.103
511