Vous êtes sur la page 1sur 5

International Journal of Advances in Science and Technology, Vol. 3, No.

4, 2011

The Detection of Amd in Human Eyes Retina Using Image Segmentation Algorithms
1

N.Udaya Kumar,2 Dr.T.RamaShri, 3 K.Srinivasa Reddy Research Scholar

Department of ECE,SriVenkateswara University College of Engineering; Tirupati,AP,India. joyudaya@yahoo.co.in Associate Professor Department of ECE, SriVenkateswara University College of Engineering; Tirupati,AP,India.

rama.jaypee@gmail.com
3

Associate Professor Department of the Electronics and Communication Engineering, NITS, Hyderabad,AP,India.
reddy_sinu2003@yahoo.co.in

Abstract
Assessment of the risk for the development of Age related Macular Degeneration requires reliable detection of retinal abnormalities that are considered as precursors of the disease. A typical sign for the later are the so called drusen, which appear as abnormal white-yellow deposits on the retina. This work presents a novel segmentation algorithm for automatic detection of abnormalities in images of the human eyes retina, acquired from a depth vision camera. Conventional image processing techniques are sensitive to nonuniform illumination and nonhomogeneous background, which obstructs the derivation of reliable results for a large set of different images. Homomorphic filtering and a multilevel variant of histogram equalization are used for non-uniform illumination compensation and enhancement. The present work develop a novel segmentation technique, the histogram based adaptive local thresholding(HALT) to detect drusen in retina images by extracting the useful information without being affected by the presence of other structures. The present work is a new algorithm for image segmentation. It use spectral histogram, which is a vector consisting of marginal distributions of responses from chosen filters as a generic feature for texture as well as intensity images. Motivated by a new segmentation energy functional and derive an iterative and deterministic approximation algorithm for segmentation. Based on the relationships between different scales and neighboring windows, it also develop an algorithm which can automatically detect homogeneous regions in an input image, which may consist of texture regions. The present work is a Bayesian grouping approach for recognition and segmentation of large-scale structures representing objects in images. It is based on detection of local image properties, extraction of simple geometrical primitives, and grouping these primitives according to probability rules and prior models. As opposed to the various template matching techniques, our method does not rely on a fixed set of input data to generate the prior with a maximum likelihood. Instead, it selects a list of subsets of the local primitives and finds the optimum set of model priors that maximizes the likelihood of the model samples representing the selected subsets. In contrast with global recognition methods that classify the whole image, the present work approach aims at solving the recognition task together with the segmentation task.

Special Issue

Page 160 of 172

ISSN 2229 5216

International Journal of Advances in Science and Technology, Vol. 3, No.4, 2011

1. INTRODUCTION
Segmentation problems are the bottleneck to achieve object extraction, object specific measurements, and fast object rendering from multi-dimensional image data. Simple segmentation techniques are based on local pixel-neighborhood classification. Such methods fail however to see global objects rather than local appearances and require often intensive operator assistance. The reason is that the logic of a object does not necessarily follow that of its local image representation. Local properties, such as textures, edgeness, and ridgeness etc. do not always represent connected features of a given object. A model-based method that sorts and groups pieces of the image primitives into objects is needed. In such a method the solution of the segmentation problem will involve resolving of the corresponding recognition task. Age-related macular degeneration (AMD) is a disease that causes progressive damage to the macula, a specialized part of the eye that allows us to see fine details clearly. When the macula malfunctions, blurring or darkness in the center of vision is experienced and tasks such as reading and driving are affected. Some common ways to detect vision loss relate to symptoms that words on a page look blurred, a dark or empty area appears in the center of vision, or straight lines are distorted. AMD is the leading cause of irreversible vision loss in people over 65 in the U.S. Although the cause of AMD is not completely understood, it has been identified that age is the greatest risk factor and there is also a hereditary nature associated with the disease. There are two forms of AMD, namely dry (also called atrophic, non-neovascular, or nonexudative) and wet (also called exudative). Dry AMD is the most common form of the disease and accounts for 90% of all AMD cases. The key identifier for dry AMD is small, round; white-yellow deposits called drusen that build up in the macula. A thorough examination by an ophthalmologist is the best way to determine if one has macular degeneration, or if he/she is at risk of developing the disease. Segmentation can be defined as a constrained partition problem. Each partitioned region should be as homogeneous as possible and neighboring regions should be as different as possible. To define a perceptually meaningful homogeneity measure, the present works use a local spectral histogram, defined as a vector of marginal distributions of responses of chosen filters as a general feature statistic for intensity as well as texture images

1.1 OBJECTIVE OF THE RESEARCH WORK


The present work for automatic detection of abnormalities in images of the human eyes retina, acquired from a depth vision camera by using a novel segmentation algorithm, an iterative and deterministic approximation algorithm for segmentation and a Bayesian grouping approach for recognition and segmentation of large-scale structures representing objects in images.

2. METHODOLOGY
Since the correct enhancement and segmentation of retina images provides valuable aid to ophthalmologists, a complete algorithm for automatic segmentation and detection of drusen has been designed and implemented in this work. Both the enhancement and thresholding operators for drusen segmentation are adaptive and based on histogram analysis. In particular, the local thresholding operator is designed from a rigorous analysis of the local histogram, as to exploit the important characteristics of the signal distribution that differentiate drusen from background and overcome the inefficiencies of other histogram based segmentation schemes. In this work, extend the model using the spectral histogram and the associated distance measure. It develop an algorithm which couples the feature detection and segmentation steps together by extracting features based on the currently available segmentation result. It also develops an algorithm which identifies regional

Special Issue

Page 161 of 172

ISSN 2229 5216

International Journal of Advances in Science and Technology, Vol. 3, No.4, 2011 features in homogeneous texture regions automatically and one for boundary localization. The main algorithmic steps are shown at Fig. 2.

Fig. 2

Algorithm for drusen detection

In technical terms, this work approach resembles the general idea behind Bayesian deformable model matching. As an example we consider the task of segmenting elongated structures for which a deformable model approach has been developed in terms of an active snake model. This class of approaches uses a description of the object model (prior) and tries to achieve the best fit with the data. Applied with many model descriptions it can solve the recognition task as well by means of maximum likelihood comparison. One deficiency of deformable model methods is their sensitivity to initialization. As a common practice, human operators are initializing the seeds at the proper locations. This means that essentially the operator is doing the bulk of the recognition task leaving only the fine-tuning to the algorithm. 3. RESULTS Two representative examples of retinal images are shown at Fig. 3. The first contains large drusen dominating extensive areas and the second one contains few, small and vaguely defined drusen. In order to demonstrate the performance of our algorithm and the efficiency of the HALT operator we proceed in a comparison with Otsus thresholding scheme over local-region histograms. Both example images are enhanced using multilevel histogram equalization (MLE) and then thresholded using Otsus and HALT techniques. A median filter is applied afterwards to remove isolated pixels. Otsus localized thresholding scheme works fine in regions that are dominated by drusen (brighter areas), since the distinction between them and the background is evident.

( a)

(b)

Fig. 3 Images after MLE enhancement, local thresholding using Otsus method at each block and median filtering to eliminate sparse pixels

This is demonstrated at Fig. 3(a), where drusen at the central part of the image are correctly distinguished from the surrounding areas. However, the algorithm is strongly affected by regions that do not contain any abnormality, like those regions at the sides of the image. Due to remaining effects of non-uniform illumination,

Special Issue

Page 162 of 172

ISSN 2229 5216

International Journal of Advances in Science and Technology, Vol. 3, No.4, 2011 parts of these regions are brighter and are misclassified as anomalies. Fig. 3-(b) brings out another disadvantage of this scheme. Vaguely defined drusen, which are either small or located inside bright background regions, are not segmented. The algorithm detects the most obvious drusen (two of them are easily conceived), but fails to detect hidden anomalies.

(a)
Fig. 3.1 pixels

(b)

Images after MLE enhancement, local thresholding using HALT method and median filtering to eliminate sparse

On the contrary, the HALT technique removes most of the background in both cases, as shown at Fig. 3.1. Even the most hard-to-see drusen are segmented without loosing their actual size and shape. Some false negatives generated by the existence of noise can be easily removed at a following detection step.

4. CONCLUSION
The paper considers histogram-based techniques for the problem of automatic AMD evaluation. The detection of anomalies in human eyes retina is a biomedical problem, appropriate for image processing and automated segmentation, whose solution is intended to help the doctors in their decision making process. Use of the proposed detector may reduce false negatives and give reliable detection accuracy in both position and mass size. We propose a histogram-based enhancement technique (MLE), which uses histogram equalization as its core operator and a histogram-based segmentation technique (HALT) to segment areas that differ slightly from their background regions. Furthermore, we establish an unsupervised and non-parametric method for drusen extraction and consider its effectiveness through several examples.

REFERENCES: [1] Otsu Nobuyuki, A Threshold Selection Method from gray level Histograms, vol. SMC-9, no.1, January 1979 [2] I. Pitas, A.N. Venetsanopoulos, Nonlinear Digital Filters1990 [3] Gonzalez R.C., Woods R.E., Digital Image Processing,1993 [4] Pratt W.K., Digital Image Processing, 1991 [5] G.Deng, L. W. Cahill and G.R. Tobin, The Study of Logarithmic Image Processing Model and Its Application to Image Enhancement [6] Joongho Chang, Gunhee Han, Hose M. Valverde,Norman C. Grisworld, J. Francisco Duque-Carillo, Edgar anchez-Sinencio, Cork Quality Classification System using a Unified Image Processing and Fuzzy-Neural Network Methodology, vol. 8, no. 4, July 1997 [7 ]Crabb D.P., Edgar D.F., Fitzke F.W., Mcnaught A.I. & Wynn H.P. (1995). New Approach to Estimating Variability in Visual-Field Data Using An Image-Processing Technique. British Journal of Ophthalmology 79:213-217.

Special Issue

Page 163 of 172

ISSN 2229 5216

International Journal of Advances in Science and Technology, Vol. 3, No.4, 2011 [8 ]Crabb D.P., Fitzke F.W., Mcnaught A.I., Edgar D.F. & Hitchings R.A. (1997). Improving the prediction of visual field progression in glaucoma using spatial processing. Ophthalmology 104:517-524. [9]Crossland M.D., Culham L.E. & Rubin G.S. (2004). Fixation stability and reading speed in patients with newly developed macular disease. Ophthalmic and Physiological Optics 24:327-333. [10]Crossland M.D. & Rubin G.S. (2006). Eye movements and reading in macular disease: Further support for the shrinking perceptual span hypothesis. Vision Research 46:590-597.

Authors Profile:

N.Udaya Kumar is Research Scholar in Electronics and Communication Engineering, S.V University, Thirupati. He received his B.Tech degree in Electronics and Communication Engineering from srinivasa engineering college, chittor and M.Tech degree in Embedded Systems from JNT University, Anathapur.

Dr.T.Rama Shri, is working as Associate Professor in EEE Department S.V University, Thirupati. She is having more than 16 years Teaching Experience. She published ten International conferences and three National Conference and interested in Digital Image Processing. Guest lecturers delivered membership of Professional bodies 1. Guest lecturers delivered 2.Life member of ISTE, Life member of IETE

K. Srinivasa Reddy is Associate Professor and Head of the Electronics and Communication Engineering, NITS, Hyderabad .He received his B.Tech degree in Electronics and Communication Engineering from JNT University, Hyderabad, and M.Tech degree in Embedded Systems from JNT University, Hyderabad. His areas of interest are Communication Systems, Cellular and Mobile Communication, Wireless Communication and Embedded Systems. He is a member of The International Association of Engineers (IAENG).

Special Issue

Page 164 of 172

ISSN 2229 5216