05-07-2013, 03:13 PM
UNSUPERVISED TECHNIQUES OF SEGMENTATION ON TEXTURE IMAGES: A COMPARISON
UNSUPERVISED TECHNIQUES.pdf (Size: 417.33 KB / Downloads: 22)
Abstract:
Unsupervised Techniques of segmentation are
simple and result in satisfactory segmented image. This
paper presents an automatic segmentation method based
on unsupervised segmentation done on Ultrasound (US)
images received from radiologist. US imaging is widely
used in clinical diagnosis and image-guided interventions,
but suffers from poor quality. One of the most important
problems in image processing and analysis is
segmentation. US image are difficult to segment due to low
contrast and strong speckle noise. Here we present three
unsupervised techniques namely thresholding, K-means
clustering and expectation maximization and compare
their results. Uniqueness of this paper is that EM
technique is used on texture featured image which gives
far better result of segmentation.
INTRODUCTION
In recent years, considerable efforts have been made in
computer-aided diagnosis (CAD) of medical images which
gives doctors and researchers a platform for detail analysis of
these images. Segmentation is the foremost step for medical
image analysis. Segmentation is a process of dividing an
image into regions having similar properties such as gray
level, color, texture, brightness, contrast etc. The techniques
available for segmentation of images can be broadly classified
into two classes: (I) based on gray level – the methods used
here are (a) amplitude segmentation methods based on
histogram features (b) edge-based segmentation © region
based segmentation and (II) based on textural feature. For
some typical applications, particularly in the medical image
processing, segmentation based on gray level does not give the
desired results; in such applications, segmentation based on
textural feature methods gives more reliable results; therefore,
texture-based analysis is extensively used in analysis of
medical images [1].
REVIEW OF TECHNIQUES USED
2-D Gabor filter is a popular tool in medical image
classification, texture analysis and discrimination. Multichannel
filtering is an excellent method for texture
investigation [5]. By processing the image using multiple
resolution techniques, it is decomposed into appropriate
texture features that can be used to classify the textures
accordingly. The multi-channel filtering approach is actually a
multiresolution decomposition process, which is similar to the
wavelet analysis. Actually a very famous class of functions
that are known to achieve both spatial and spatial frequency
localization is the Gabor function that is not truly a wavelet (in
the mathematical sense) but it can be implemented in such a
manner as to mimic the properties of a wavelet.
The process of texture segmentation using Gabor filters
involves a proper filter bank design that should be tuned to
different spatial-frequencies and orientations to cover the
spatial frequency space, decomposing the image into a number
of filtered images; feature extraction from these images, and
clustering of the pixels in the feature space to produce
segmented image. On generating texture features using multichannel
filters two primary issues must be addressed. The first
issue deals with the functional characterization of the channels
as well as their number, orientation and spacing. The second
issue deals with extracting significant features by data
integration from different channels. Texture segmentation
requires simultaneous measurements in both the spatial and
the spatial frequency domains. Filters with smaller bandwidths
in the spatial frequency domain are more desirable as they
allow us to do finer distinctions. On the other hand, accurate
localization of texture boundaries requires filters that are
localized in the spatial domain.
PROPOSED METHOD
In our work, we propose for segmentation of ultrasound
image, three unsupervised segmentation techniques namely
image segmentation through k–means clustering algorithm,
segmentation using thresholding and image segmentation
using Expectation Maximization (EM) Algorithm. All these
techniques were used on texture featured US image. EM
method of segmentation done after feature extraction by
Gabor filters gives good segmentation results.