INTRODUCTION
For some applications, such as image recognition or compression, we cannot
process the whole image directly for the reason that it is inefficient and unpractical.
Therefore, several image segmentation algorithms were proposed to segment an
image before recognition or compression. Image segmentation is to classify or
cluster an image into several parts (regions) according to the feature of image,
for example, the pixel value or the frequency response (Huang
and Wu, 2006).
Neurological conditions are the most common cause of serious disabilities and
have a major, but often unrecognized, impact on health and social services.
It can change the shape, volume and distribution of brain tissue. The advantages
of Magnetic Resonance Imaging (MRI) over other diagnostic image modes are its
high spatial resolution and excellent discrimination of soft tissues. MRI is
the preferred imaging techniques for examining neurological conditions which
requires segmentation into different classes which are regarded as the best
available representations for biological tissues and it can be performed by
using image segmentation (Birgani et al., 2008).
Spectral pattern is not sufficient in high resolution image for image segmentation
due to variability of spectral and structural information. Thus, the spatial
pattern or texture techniques are used. Thus, the concepts of Holder Exponent
for segmentation in high resolution image are used. Holder Exponent is basically
used to measure the local regularity of image (Trujillo
et al., 2012).
Local binary patterns is a technique that describes the texture in terms of
both statistical and structural characteristics. Holder exponent is used to
assess the roughness or smoothness around each pixel of the image. The measure
of dispersion is used to compute the Holder Exponent (Pietikainen
et al., 2011).
The window size is assessed to detect the localized singularities. Larger window
size is insensitive to noise that leads to loss of information of singularity,
while the smaller window size represent the singularity well but sensitive to
noise. So, it is preferable to determine the window from two respects:
• 
Additional singularity should not be contained in the same
window 
• 
The size of the window should be enlarged on the location without obvious
singularity 
An iterative clustering procedure is adapted to detect the range of cluster
contained in the kernel, localize the cluster center (this approach moves the
range of holder exponent values in the direction where the density is higher)
and identify the cluster contained in the kernel (background, range). A clustering
procedure including maximum likelihood analysis is used to classify the Holder
Exponent image.
RECENT STATUS OF IMAGE SEGMENTATION
Incorporating local spatial and gray information together (Cai
et al., 2007), a novel fast and robust FCM framework for image segmentation,
i.e., Fast Generalized Fuzzy cmeans Clustering Algorithms (FGFCM). FGFCM can
mitigate the disadvantages of FCM_S and at the same time enhances the clustering
performance.
A new approach regarding matting problem (Rhemann et
al., 2008) which splits the task into two steps: Interactive trimap
extraction followed by trimapbased alpha matting. That paper has two contributions:
(1) A new trimap segmentation method using parametric maxflow and (2) An alpha
matting technique for high resolution images with a new gradient preserving
prior on alpha.
Image segmentation can be performed on raw radiometric data, but also on transformed
colour spaces, or, for highresolution images, on textural features. TriasSanz
et al. (2008) have reviewed several existing colour space transformations
and textural features and investigate which combination of inputs gives best
results for the task of segmenting highresolution multispectral aerial images
of rural areas into its constituent cartographic objects such as fields, orchards,
forests, or lakes, with a hierarchical segmentation algorithm.
A threedimensional atlas of the mouse brain (Dorr et
al., 2008) manually segmented into 62 structures, based on an average
of 32 μm isotropic resolution T2weighted, within skull images of forty
12 week old C57Bl/6J mice, scanned on a 7 Tscanner. Individual scans were normalized,
registered and averaged into one volume. Structures within the cerebrum, cerebellum
and brainstem were painted on each slice of the average MR image while using
simultaneous viewing of the coronal, sagittal and horizontal orientations.
An optimization approach is proposed to minimize overand under segmentations
in order to attain more accurate segmentation results using Definiens Developer
software (Esch et al., 2008). The optimization
iteratively combines a sequence of multiscale segmentation, featurebased classification
and classificationbased object refinement. The developed method has been applied
to various remotely sensed data and was compared to the results achieved with
the established segmentation procedures.
A new and fast unsupervised technique for segmentation of highresolution Synthetic
Aperture Radar (SAR) images (Galland et al., 2009)
into homogeneous regions. That technique was based on Fisher probability density
functions of the intensity fluctuations and on an image model that consists
of a patchwork of homogeneous regions with polygonal boundaries.
Texture in highresolution satellite images requires substantial amendment
in the conventional segmentation algorithms. Chakraborty
et al. (2009) proposed a measure to compute the Holder Exponent (HE)
to assess the roughness or smoothness around each pixel of the image. The localized
singularity information was incorporated in computing the HE.
A method for detecting the high resolution locations of membranes from low
depthresolution images (Glasner et al., 2011).
They have approached that problem using both a method that learns a discriminative;
overcomplete dictionary and a kernel SVM.
PROPOSED WORK
Here, an efficient image segmentation technique to segment the high resolution
medical images are proposed. Initially, the filtering technique is applied to
the query image to remove the noise content in the medical image. Then morphological
operations like dilation and erosion are done over the filtered image. Finally,
the image is segmented using Holder exponent.
The basic flow diagram of the proposed method as shown in the Fig.
1.
Gabor filtering: Gabor filters can serve as excellent bandpass filters
for onedimensional signals. A Gabor filter is a linear filter whose impulse
response is defined by a harmonic function multiplied by a Gaussian function.
Gabor filters is having various transforms, image properties, operators, frequencies
and various features which are used in detecting image segmentation. Gabor filter,
a kind of frequency filter, which has been applied to texture analysis, moving
object tracking and face recognition, are also shown to be good fits in character
recognition field. The primary step for high resolution image segmentation is,
removing the noise from the query image using Gabor filter. This filter removes
the noise content from the image and makes the image ready for the recognition.
The complex term of the image g (x, y) can be represented as:

Fig. 1: 
Basic flow of the proposed system 
The real component of the image g (x, y) can be represented as:
The imaginary component of the image can be represented as:
Where:
In this equation, λ represents the wavelength of the sinusoidal factor,
θ represents the orientation of the normal to the parallel stripes of a
Gabor function, ψ is the phase offset, σ is the sigma of the Gaussian
envelope and γ is the spatial aspect ratio and specifies the ellipticity
of the support of the Gabor function.
Morphological operations: Morphology is a broad set of image processing
operations that process images based on shapes. Morphological operations apply
a structuring element to an input image, creating an output image of the same
size.
Morphological operations are affecting the form, structure or shape of an object.
Applied on binary images (black and white imagesImages with only 2 colors:
Black and white). They are used in pre or post processing (filtering, thinning
and pruning) or for getting a representation or description of the shape of
objects/regions (boundaries, skeletons convex hulls). In a morphological operation,
the value of each pixel in the output image is based on a comparison of the
corresponding pixel in the input image with its neighbors. By choosing the size
and shape of the neighborhood, you can construct a morphological operation that
is sensitive to specific shapes in the input image.
The two principal morphological operations are dilation and erosion. Dilation
allows objects to expand, thus potentially filling in small holes and connecting
disjoint objects. Erosion shrinks objects by etching away (eroding) their boundaries.
These operations can be customized for an application by the proper selection
of the structuring element, which determines exactly how the objects will be
dilated or eroded. The dilation process is performed by laying the structuring
element B on the image A and sliding it across the image in a manner similar
to convolution (will be presented in a next laboratory). The difference is in
the operation performed. It is best described in a sequence of steps:
Step 1: 
If the origin of the structuring element coincides with a
'white' pixel in the image, there is no change; move to the next pixel 
Step 2: 
If the origin of the structuring element coincides with a 'black' in image,
make black all pixels from the image covered by the structuring element 
The erosion process is similar to dilation, but we turn pixels to 'white',
not 'black'. As before, slide the structuring element across the image and then
follow these steps:
Step 1: 
If the origin of the structuring element coincides with a
'white' pixel in the image, there is no change; move to the next pixel 
Step 2: 
If the origin of the structuring element coincides with a 'black' pixel
in the image and at least one of the 'black' pixels in the structuring element
falls over a white pixel in the image, then change the 'black' pixel in
the image (corresponding to the position on which the center of the structuring
element falls) from ‘black’
to a 'white' 
Image Transformation using HE: The Holder exponent analysis is used
here to transform the image for the identification of the texture. It does not
require any prior information about the pixel intensity. The predefined measure
is used to estimate the degree of texture around each pixel. The predefined
measure is one of the most important characteristics to compute the Holder exponent.
The roughness or smoothness around each pixel can be assessed by the appropriate
estimation of the measure. In this paper the measure of dispersion of pixel
values using linear regression analysis are determined.
Let the subset Ω* of the region Ω contains only those pixels which
intersect the perimeter of the circle of radius r. Hence, for t number of increasing
radius (i.e., r = 1 to t ) there will be t number of subsets Ω*. Subsequently,
the radius (r) versus the intensity values I (i) of that subset Ω* is plotted.
From the least square fit of regression line, the intensity value J is calculated
for each radius (r). As a result, a new measure K (I) = I(i)J, for each I∈Ω*
is obtained. In turn this provides the dispersion of pixels from the line of
regression. The above measure can be represented as:
where, J is the derived intensity value for radius r using the regression equation
and μdisp (Ω*) is the measure of dispersion of pixels contained in
the subset Ω*.
Logarithmic plots of computed measure K versus radius R values are drawn and
got the Holder exponent α as follows:
where, t is the total number of identified balls, m is the number of intersected
pixel on the perimeter of the circle of radius R (r) and N is the total number
of pixels under each ball of radius R (r).
Clustering: The range RQ of a cluster in the Holder exponent image is
defined as follows Let us consider the below equation:
G = {gkl, Holder exponent value in G (k, l)} 
where, k = 1, ..., m and l = 1, ..., m is a kernel with m^{2} Holder
Exponent (HE). Q is a cluster in G with center GQ (mean). Then, the range RQ
of the cluster Q contains only those HE values satisfying the following properties:
Equation 8 means that cluster Q contains that range of HE
value, which have a minimum degree of association (represented by RQ).
Localization of cluster is to find a center in the dataset where the ‘density’
(or number) of range of pixel values in G within a range, i.e., RQ is locally
maximal. Primarily the cluster center is initialized with the mean HE values.
Then, the selection of HE values within the RQ from the center of G (i.e., mean
of G).
This is implemented iteratively by decreasing RQ with a constant value until
absolute difference between the initial center (CQ) and present center (ME)
reaches the desired value (minimum difference). In the first iterations (when
RQ is still large) this technique moves the range of HE values to regions of
the data where the ‘global’
density is higher (these regions often contain the large number of pixels).
After some iteration (when RQ is equal to constant value) the kernel center
moves towards an actual range of HE values where the density is ‘locally’
higher.
The cluster identification consists of two parts, Background and Range. Backgrounds
are the HE values in the HE image not included between (CQRQ) and (CQ+RQ) values.
Such HE values, either belongs to another cluster or do not belong to any cluster
(noise, are not significantly associated with other HE values). HE values belonging
to other clusters are not considered at the time of threshold calculation for
the current cluster. Ranges are the HE values represented as (CQRQ)≤HE≤(CQ+RQ).
HE values belonging to the cluster are significantly correlated.
Cluster weight is computed with the formula:
where, k is the number of cluster resides in the kernel. F_{req} is
the total number of HE falling in the range of kth cluster residing in the kernel.
W (cluster_{k}) is the possibility (or weighting factor) to assign the
HE value in the kth cluster. n and m represent the number of row and column
of the kernel, respectively. Maximum weighted cluster is identified with the
equation:
where, L is the number of cluster contained in the kernel.
RESULTS AND DISCUSSION
To show the robustness of the proposed HE analysis is used in this study. This
is done by applying the Kmean separately on the transformed images obtained
from the Holder exponent analysis and comparing the results with other algorithms.
Simultaneously the results of the proposed segmentation method, compared with
other results to show the robustness of the proposed segmentation method.
The results obtained during the process of proposed medical image segmentation
are discussed. Initially, Gabor filter has to be applied to the input query
image to reduce the noise content in the image. Since, the segmentation has
to be done in a clear image to get accurate segmented output. Figure
2 shows the query image and also the image output of Gabor filter.
After applying Gabor filter, the output image is subjected to morphological
operations like dilation and erosion. Figure 3 shows the output
image after morphological operations.
Then the image is transformed using Holder exponent. Holder Exponent is used
to assess the roughness or smoothness around each pixel of the image. The measure
of dispersion is used to compute the Holder Exponent.

Fig. 2(ab): 
(a) Input image and (b) Gabor filter output 

Fig. 3(ab): 
(a) Input image and (b) Image after morphological operations 
After image transformation, clustering is applied to cluster the image contents
to form the segmented image. This noise can be removed by applying the mean
value for each pixel from the neighbor pixels. Thus the segmentation output
of the given medical image as shown in the Fig. 4.
The other algorithms needs number of class as an input to segment the image
whereas proposed technique need not require any input for segmentation. They
only considers the information of pixel value for segmentation. The proposed
segmentation technique also considers association of HE values to increase the
classification accuracy.
In other methods, an iteration procedure is carried out until all pixels get
classified in the image whereas in the proposed technique, iteration is done
to identify the range of HE densely occupied in the kernel, and to partition
those Holder exponent into a cluster which matches with that range. Holder exponent
values (noise or not associated with the other cluster) are clubbed into a nearest
possible cluster using the local maximum likelihood analysis. Proposed segmentation
method is simple and takes very less computation time while other algorithms
is time consuming.
Comparative analysis: Magnetic Resonance Imaging (MRI) is one of the
most common ways to visualize brain structures. Based on this imaging technique,
the study of the main cerebral tissues (namely, White Matter (WM) and Grey Matter
(GM)) is in particular a key point in the context of computeraided diagnosis
and patient followup. Our proposed image segmentation technique is compared
with the existing technique depend upon the white matter and grey matter of
the segmented brain MRI image.
Table 1 show the percentage of white matter and grey matters
in the proposed image segmentation as well as the existing segmentation technique.

Fig. 4(ab): 
(a) Input image and (b) Segmented image 
Table 1: 
Overlap measures (GM, WM) obtained for different segmentation
methods 

CONCLUSION AND FUTURE SCOPE
Image segmentation is the most challenging and active research area in the
field of image processing for the last decade. In spite of the availability
of a large variety of stateof art methods for brain MRI segmentation, but still,
brain MRI segmentation is a challenging task and there is a need and huge scope
for future research to improve the accuracy, precision and speed of segmentation
methods. Here the medical image segmentation algorithms based on Holder exponent
are proposed. Since, HE can be used as a tool to measure the roughness or smoothness
around each pixel in the image and also HE does not require any prior information
about the pixel intensity. Present work gives more overlap measures as compared
to the existing technique, thus, our medical image segmentation technique is
more efficient. The proposed segmentation results shows that, the use of Holder
exponent based strategy globally leads to better results than the other state
of the methods existing now.
The availability of very high spatial resolution images in remote sensing brings
the texture segmentation of images to a higher level of complexity. Such images
have so many details that the classical segmentation algorithms fail to achieve
good results. In the case of BRAIN images, a texture can be so different within
a same class that it becomes very difficult even for a human to segment or interpret
those images. The study of the high frequency content of the data seems to be
a good way to study those images. A new method which uses the singularity information
to achieve the segmentation. It is based on the computation of the Holder regularity
exponent at each point in the image.