International Journal of Electrical and Computer Engineering (IJECE) Vol. No. June 2014, pp. ISSN: 2088-8708 Retinal Blood Vessels Extraction Based on Curvelet Transform and by Combining Bothat and Tophat Morphology Gayathri. Narmadha. Thilagavathi. Pavithra. Pradeepa Departement of Electronics and Communication Engineering. Abdul Hakeem College of Engineering and Technology Vellore. India Article Info ABSTRACT Article history: Retinal image contains vital information about the health of the sensory part of the visual system. Extracting these features is the first and most important step to analysis of retinal images for various applications of medical or human recognition. The proposed method consists of preprocessing, contrast enhancement and blood vessels extraction stages. In preprocessing, since the green channel from the coloured retinal images has the highest contrast between the subbands so the green component is selected. To uniform the brightness of image adaptive histogram equalization is used since it provides an image with a uniformed, darker background and brighter grey level of the blood vessels. Furthermore Curvelet transforms is used to enhance the contrast of an image by highlighting its edges in various scales and Eventually the combination of Bothat and Tophat morpholological function followed by local thresholding is provided to classify the blood vessels. Hence the retinal blood vessels are separated from the background image. Received Mar 12, 2014 Revised May 6, 2014 Accepted May 25, 2014 Keyword: Blood vessels extraction Curvelet transform Morphological function Preprocessing Retinal image Copyright A 2014 Institute of Advanced Engineering and Science. All rights reserved. Corresponding Author: Gayathri Departement of Electronics and Communication Engineering. Abdul Hakeem College of Engineering and Technology. Vellore. India. Email: kgayathri. ece@gmail. INTRODUCTION One of the most important internal components of eye is called retina, which covers all posterior Any damage in retina leads to severe diseases. Disorders in retina resulted from special diseases are diagnosed by special images which are obtained by using optic imaging called Fundus image. The Fundus images are used for diagnosis by trained clinicians to check for any abnormalities or any change in the retina. They are captured by using special devices called ophthalmoscopes. Each pixel in the fundus image consists of three values namely Red. Green and Blue, each value being quantized to 256 levels. The blood vessels are the important parts of the retinal images consisting of arteries and arterioles. Checking the obtained changes in retinal images in an especial period can help the physician to diagnose the disease. Applications of retinal images are diagnosing the progress of some cardiovascular diseases, diagnosing the region with no blood vessels (Macul. , using such images in biometric applications and in helping automatic laser surgery on eye, etc. On the other hand, extracting the retinal blood vessels is done in some cases by physician manually, which is difficult and time consuming and is accompanied by high mistakes due to much dependence on the physicians skill level. So the exact extraction of the blood vessels from the retinal images necessitates using algorithm and instruments which reduce the dependency on the function and eliminate the error factors. While capturing the image because of the variability of light reflection coefficient in different parts of the retinal layer also due to the defects in imaging systems there occurs nonuniform illumination in the retinal image, pixels related to the blood vessels cannot be classified carefully. This improper contrast is due to Journal homepage: http://iaesjournal. com/online/index. php/IJECE A ISSN: 2088-8708 different vessels have different contrast. arteries have higher contrast than veins. In addition to this, presence of noise, fovea and optical disk, width of the vessels, effects of lesions and pathological changes should also be considered. So, for extraction of blood vessels with high accuracy, we need of an effective algorithm. In the proposed algorithm, the focus will be on the extraction of blood vessels constitutes of digital colored images of retina as its input which is then converted to green channel image with best contrast. Since the preprocessing phase plays an important role in final extraction results. One of the advantages of this phase is by applying the adaptive histogram equalization and Curvelet transform on the image to reduce the noise and improve the contrast. Therefore the inadequacy of previous methods is resolved. Since the blood vessels are distributed in different directions, applying morphological operation causes the blood vessels with high accuracy to be separated from the background and finally the connected components with defined threshold, frills in the image are removed and extracted blood vessels are obtained. LITERATURE REVIEW Recently many automated detection techniques are constantly devised and implemented to help ophthalmologists detect blood vessels by applying image processing and pattern recognition techniques. In 2012. Kalaivani. Jeyalakshmi and Aparna. V . used Adaptive Histogram Equalization for initial enhancement, followed by this the curvelet transforms to the equalized image and the curvelet coefficients are obtained. The vessel extraction is done based on thresholding technique and the KirschAos It involves spatial filtering of the image using the templates in eight different orientations. The masking of redundant regions in the obtained output image is carried out using boundary techniques. In other related work. Marwan D. Saleh and C. Eswaran . proposed the algorithm has employed techniques, such as background removal, contrast enhancement, h-maxima transformation, thresholding, etc. After converting the RGB image to gray-scale, both morphological top-hat and bottom-hat transforms have been exploited to perform the contrast enhancement. Other techniques such as h-maxima transform and multilevel thresholding have been exploited to decrease the intensity levels as much as possible to facilitate the threshold selection for binarization in 2012. Iqbal. I et al . in 2007 used Color Space Conversion. Edge Zero Padding. Median Filtering and Adaptive Histogram Equalization as pre-processing techniques and they used segmentation to group the image into regions with same property or characteristics. Methods of image segmentation include simple thresholding. K-means Algorithm and Fuzzy C-means. Since it takes more time to load the data. An efficient retinal analysis method based on curvelet transform and multi structure elements was proposed by Miri et al . in 2011, he described that green channel of the original colored image was Obtain the fundus region mask using Otsu algorithm followed by morphological closing and multiply its result image with FDCT via wrapping, then modify the curvelet coefficients and obtained the enhanced image. Then subtracts the estimated background from the enhanced image. Thereby modified tophat transforms using the multistructure elements morphology were applied and by providing opening function the image were reconstructed. In order to eliminate the remained false edges, apply length filtering along with CCA . locally but the image resulted from TopHat function can include all negligible changes in the grey levels existing in the image . uch as nois. Priya R et al . in 2011 used preprocessing techniques like Gray scale Conversion. Adaptive Histogram Equalization. Matched Filter Response and proposed a method for feature extraction based on Area of on pixels. Mean and Standard Deviation. Also in 2012. Jaspreet Kaur and Dr. Sinha . presented a Filter based approach with morphological filters is used to segment the vessels. The morphological filter is tuned to match that part of vessel to be extracted in a green channel image. To classify the pixels into vessels and non vessels local thresholding based on gray level co-occurrence matrix as it contained information on the distribution of gray level frequency and edge information have been presented. In 2012. Paintamilselvi et. carried out blood vessels extraction in five steps. First the RGB image was converted into gray scale. Secondly morphological opening and closing operation is used to reduce small noise. In the third step to obtain the vessel structure a unique technique called top hat transformation was used. In the fourth step, the resultant image was obtained after binarisation and Finally connected component analysis was used to obtain an image which was free from noise. The rest of the paper is organized as follows: In Section 3 proposed methods is described while 1 & 3. 2 examines green channel selection and image enhancement using adaptive histogram equalization. In Section 3. describes Contrast enhancement using FDCT and section 3. 2 presents the method for extraction of vessels from colored retinal image. In section 4 the results of the algorithm over an extensive dataset are presented and conclusions are reviewed in section 5. IJECE Vol. No. June 2014 : 389 Ae 397 IJECE ISSN: 2088-8708 PROPOSED METHOD The proposed system in this work consists of following steps preprocessing and blood vessels The block diagram of retinal blood vessels extraction is shown in Figure 1. Preprocessing Input RGB Image Green channel selection Adaptive histogram equalization Contrast enhancement using FDCT Blood vessel Edge detection using morphological operation followed by local Blood vessels extracted image Figure 1. Block diagram of retinal blood vessels extraction technique Preprocessing Green Channel Selection If the three channels of a RGB coloured retinal image are observed, the red channel shows a poorly contrasted retinal vasculature on top of the choroidal vasculature. The Green channel shows well contrasted arteries and veins with a clear dark fovea in the centre. The blue channel shows a noisier image of the So that green channel has the best contrast by experience is shown in Figure 2. Hence it is selected for further work. Figure 2. Red channel, . green channel and . blue channel Adaptive Histogram Equalization We initially worked on the colour retinal image. To reduce the effect of different lightning conditions and to uniform illumination Adaptive histogram Equalization is used. It is an enhancement technique capable of increasing the local Contrast also it improves the brightness of an image. It differs from Retinal Blood Vessels Extraction Based on Curvelet Transform and by Combining Bothat A (K. Gayathr. A ISSN: 2088-8708 ordinary histogram equalization in the respective that adaptive method computes several histograms each corresponding to distinct section of the image and uses them to redistribute the lightness values of image. that contrast of the image was adjusted to the limit 0 and 1 hence the blood vessels are highlighted. Contrast Enhancement using Fast Discrete Curvelet Transform Curvelet transform is developed to overcome the limitation of wavelet and Gabor transforms . Although, wavelets are widely used in feature extraction but it fails to handle randomly oriented edges of the object and the singularities of the object. Gabor filters overcome the limitation of wavelet transform and deal with the oriented edges, but it loses the spectral information of the image. Curvelet transform is used to overcome these problems of the wavelet and Gabor filters. It can obtain the complete spectral information of the image and handle with the different orientations of the image edges. The idea of curvelet is to represent a curve as a superposition of functions of various lengths and widths obeying the scaling law width OO length2. This can be done by first decomposing the image into subbands i. separating the object into a series of disjoint scales. Then, each scale is analyzed by a local ridgelet transform. The newly constructed and improved version of the curvelet transform is known as Fast Discrete Curvelet Transform (FDCT). The new constructed version is faster, simpler and less redundant than the original curvelet transform, which based on Ridgelet. As mentioned, according to Cand'es et al. two implementations of FDCT are proposed: Unequally spaced Fast Fourier Transform (USFFT) Wrapping Function Both implementations of FDCT differ mainly in choosing the spatial grid that used to translate curvelet at each scale and angle. Both digital transformations return a table of digital curvelet coefficients indexed by scale, orientation and location parameters. Here, we use the wrapping method to implement the Fast Discrete Curvelet Transform (FDCT) on the retinal image which is a two dimensional signal. The wrapping implementation is simpler, faster and has less computational complexity than existing approaches. Wrapping based curvelet transform is a multi-scale pyramid which consists of different orientations and positions at a low frequency level. Basically, multiresolution discrete curvelet transform in the spectral domain utilizes the advantages of fast Fourier Transform (FFT). During FFT, both the image and the curvelet at a given scale and orientation are transformed into the Fourier domain. At the end of this computation process, we obtain a set of curvelet coefficients by applying inverse FFT to the spectral product. This set contains curvelet coefficients in ascending order of the scales and orientations . Figure 3. Steps in FDCT via wrapping method In order to obtain the curvelet coefficients for an image the below steps are performed sequentially. Apply the 2D FFT and obtain Fourier samples . 1,n. , -n/2