Retinal Based Image Enhancement Using ... - Semantic Scholar
Retinal Based Image Enhancement Using Contourlet Transform P. Sharath Chandra, M.C. Hanumantharaju, and M.T. Gopalakrishna Department of Information Science and Engineering, Dayananda Sagar College of Engineering, Bangalore, India {sharath.chandra174,mchanumantharaju,gopalmtm}@gmail.com
Abstract. In medical image processing, retinal image enhancement is the challenging issue to reveal the unseen details of an retinal image, thus, in many applications image enhancement issued to solve the challenges such as, noise reduction, blurring, degradation, etc. To improve the visual grade of retinal images we have many alternative image enhancement techniques that are suitable for specific application. This paper presents an overview of various retinal image enhancement techniques that will process the original Retinal image to obtain enhanced image suitable for a specific application. The method used in this paper has been evaluated with help of PSNR image Quality measure which is applied over several retinal images which is obtained from the datasets such as DRIVE, STARE and few other’s provided by local medical experts. The comparative experimental results indicate that our proposed enhanced method has better outcome. Keywords: Image Processing, Retinal Images, Image analysis, Relative study, Retinal Image Enhancement Techniques, Contourlet Transform, Multi-scale decomposition.
1
Introduction
Image enhancement is used for the refinement of image quality, and to provide detailed information for Retinal image examiner, and also for any automated image processing techniques. During this refinement, one or more attributes of an image is reformed, for a particular task. Depending on the task the attributes are modified without spoiling original image. Spatial Domain method and Frequency Domain method are the two main categories in image enhancement. During the diagnosis and detection of many Retinal diseases, image enhancement techniques are used to recognize Diabetic Retinopathy (DR) and Age Related Muscular degradation (AMD). thus, refinement of an image is mandatory for diagnosis performed either automatically or manually A Retinal image shown in Fig.1, is affected by nonuniform illumination, in the captured image it can be seen that contrast and luminosity is not uniform. Thus the fundamental problem is to improve the quality of an image © Springer International Publishing Switzerland 2015 S.C. Satapathy et al. (eds.), Proc. of the 3rd Int. Conf. on Front. of Intell. Comput. (FICTA) 2014 – Vol. 2, Advances in Intelligent Systems and Computing 328, DOI: 10.1007/978-3-319-12012-6_64
581
582
P.S. Chandra, M.C. Hanumantharaju, and M.T. Gopalakrishna
during retinal image analysis. There are various filters that are available such as Gaussian filter, median filter, High pass filter and Low pass filter that are used to remove the noise that corrupt the retinal images. The Basic pre-processing steps for image enhancement are: 1. 2.
3.
Input image: - Image can be blur images, medical image etc. Pre-processing of an image: - Before applying various image enhancement techniques on the input images, various pre-processing methods are applied on those images. Applying domain techniques: - By using domain techniques quality of Pre-processed image is enhanced[12]. Such as, Log Transformation Technique Power Law Transformation Technique Alpha rooting Technique
Fig. 1. Retinal Image with Uneven Illumination and Contrast
2
Related Work
M Emre Celebi et al[14]. have proposed the feature preserving contrast enhancement Method for Retinal vascular Images using discrete-shearlet transform (DST) and perceptual uniform color space CIEL*a*b*. The study specify that Diabetic retinopathy and age related macular degeneration are the diseases that cause the retinal impairment and blindness. This method is widely used for both color and grayscale image enhancement, however this study does not consider the uneven illumination and the noise reduction during reconstruction of the image.
Retinal Based Image Enhancement Using Contourlet Transform
583
Peng Feng at al[15]. have proposed the concept of Contourlet transform over the retinal images using multi-scale edge enhancement. This study focus on the contrast enhancement which is necessary in pre-processing step for both natural images and ophthalmic images. In this method the image enhancement is performed using Contourlet transform which comprises of the following two steps: Laplacian pyramid (LP) and directional filter bank (DFB).
3
Proposed Method
Today, the Retinal Imaging is widely used in the diagnosis of retinal diseases such as blindness and visual impairments. The images obtained through digital devices contains uneven light illumination and some amount of noise, this results in poor quality of image which is not feasible for medical diagnosis. So far many methods are used world wide to enhance the retinal images, which have marginal drawbacks when compared. In this proposed method we improve the visibility of image optimally for the detection of various retinal diseases. For experimental purpose, various input retinal images have been obtained from the datasets DRIVE, STARE and few other’s provided by local medical experts. As indicated in Fig [2]. Initially the retinal image input obtained can be a gray scale image or color image. In an input image the edges may be brighter than the background area and surrounding edges may be weaker thus Discrete-shearlet Transform(DST) is applied on input retinal image. When the input is color image then Red, Green and Blue components are extracted. Since Green component has the optimal visibility we consider only green component as a primary source for the image enhancement. Now the green component is converted into gray-scale. Further CLAHE is applied on green component gray-scale image or any suitable HE method is applied as applicable to the type of image. After performing DST, at the time of image reconstruction there will be marginal uneven illumination and noise generated. Hence we use the methods such as Gamma correction which reduces the uneven illumination to obtain optimum image quality and suitable filters to remove different kinds of noise generated. Followed by Contourlet transform is applied. Finally we verify the image quality by using peak signal to noise ratio (PSNR) quality metric. By following the above proposed architecture, finally we get the optimal enhanced retinal image which has the best PSNR quality measure as experimentally compared.
584
P.S. Chandra, M.C. Hanumantharaju, and M.T. Gopalakrishna
Fig. 2. Proposed architecture
3.1
RGB Color Components
In any given color inputs, the RGB color components are extracted. By extracting these components we can observe the image clarities in all three components, in all the images of the extracted components, the green channel is observed to have significantly better visual quality than others as shown in Fig [3.1] below.
Fig. 3. Color components of RGB Image
Retinal Based Image Enhancement Using Contourlet Transform
585
The above selected green channel component is converted to gray-scale image, this gray-scale image can be feasible to apply gamma correction in reduction of uneven illumination, and noise reduction is possible by applying suitable filters. Thus we can infer that image is ready for applying contourlet transform. In an obtained input image the RGB color component extraction plays a vital role, further where the green channel component is responsible for image enhancement in getting the optimal result. 3.2
Contourlet Transform
The method Contourlet transform is associated with the decomposition of image into smaller sectors. By Contourlet transform we obtain sector wise smoother image along with smooth contours. The Contourlet transform technique overcome the challenges over traditional methods as wavelet transform and curvelet transform techniques. In this proposed method we make use of multi-scale decomposition. Multi-scale decomposition technique is best suited to get enhanced image. According to the study, Laplacian decomposition method is widely used for Image enhancement process. There exist a few drawbacks by using Laplacian pyramid in this method. Laplacian pyramid does not focus on oscillations in the images. Since the Laplacian Pyramid give chance of missing a few sub-bands during the reconstruction process image might not have the optimal resolution. Due to the above mentioned drawbacks, in the proposed method we make use of the multi-scale decomposition where the oscillations in the image is identified and we preserve the edges . By preserving edges the image obtained is smoother. The method of multi-scale decomposition focus on improving the contrast and preserve the contours quality. This makes the retinal image clearly visible and final outcome will have the optimal value. The scaling ratio is maintained uniform and during reconstruction process the quality of image is not lost, where as in traditional methods such as Laplacian pyramid this is a challenge to overcome.
(a)
(b)
Fig. 4. Contourlet Transform: (a) Input Image (b) Enhanced output Image
586
4
P.S. Chandra, M.C. Hanumantharaju, and M.T. Gopalakrishna
Experimental Results
For the experimental purpose we consider various retinal image inputs which are obtained from datasets such as DRIVE, STARE and few other’s provided by local medical experts. The obtained inputs are either gray scale or color image. The Experiment is carried out for all kind of images, and we could observe optimal results tested under various conditions. Below Fig [4.1] shows pictorial representation of the process where we can see the optimal enhanced image after applying contourlet transform, in which the obtained enhanced image is feasible for medical diagnosis. Compared to traditional methods, this proposed experimental result provides the highest PSNR value.
Fig. 5. Experimental Result: (a) Input color image (b) Extracted Green channel of color image (c) gray-scale of green channel image.
Fig. 6. Experimental Result: (a) Gamma correction after DST (b) CLAHE (c) Enhanced Image obtained using Contourlet Transform
5
Conclusion
The proposed method focuses on improving the visibility of image based on quality metrics like Edge Improvement Index, Contrast Improvement Index and Peak signal to Noise ratio. As evidenced by experiments with the Contourlet transform, there is a better preservation of contours than with other methods. Traditional methods does not consider the uneven illumination and noise reduction, hence the proposed method takes care of both to provide optimal image quality which is very appropriate for medical diagnosis. Finally proposed method tries to maintain a balance between PSNR, EII and CII, which will give us all these values in the accepted range and the comparative experimental results indicate that our proposed enhanced method has better outcome.
Retinal Based Image Enhancement Using Contourlet Transform
587
References 1. Jamal, I., Akram, M.U., Tariq, A.: Retinal Image Preprocessing: Background and Noise Segmentation. TELKOMNIKA 10(3), 537–544 (2012) 2. Bedi, S.S.: Various Image Enhancement Techniques- A Critical Review. International Journal of Advanced Research in Computer and Communication Engineering 2(3) (2013) 3. Joshi, G.D., Sivaswamy, J.: Colour Retinal Image Enhancement based on Domain Knowledge. In: Sixth Indian Conference on Computer Vision, Graphics and Image Processing (2008) 4. Shaeidi, A.: An Algorithm for Identification of Retinal Microaneurysms. Journal of the Serbian Society for Computational Mechanics 4(1) (2010) 5. Wang, Y., Tian, D., Vetro, A.: A Local Depth Image Enhancement Scheme for View Synthesis. Copyright Mitsubishi Electric Research Laboratories, Inc. (2012) 6. Maini, R., Aggarwal, H.: A Comprehensive Review of Image Enhancement Techniques. Journal of Computing 2(3) (2010) ISSN 2151-9617 7. Yahya, S.R.: Review on Image Enhancement Methods of Old Manuscript with Damaged Background. International Journal on Electrical Engineering and Informatics 2(1) (2010) 8. Vij, K., Singh, Y.: Enhancement of Images Using Histogram Processing Techniques. Int. J. Comp. Tech. Appl. 2(2), 309–313 (2009) 9. Kaur, N.: Image Enhancement Techniques a Selected Review. International Journal of Engineering Research and Technology (IJERT) 2(3) (March 2013) 10. Robinson, P.E., Lau, W.J.: Adaptive Multi-Scale Retinex algorithm for contrast enhancement of real world scenes (2012) 11. Kaur, E.M., Jain, E.K., Lather, E.V.: Study of Image Enhancement Techniques: A Review. International Journal of Advanced Research in Computer Science and Software Engineering 3(4) (2013) 12. Gavet, Y., Fernandes, M., Pinoli, J.-C.: Quantitative evaluation of image registration techniques in the case of retinal images. @JEIGavet 2012 (2012) 13. Mundhada, S.O., Shandilya, V.K.: Spatial and Transformation Domain Techniques for Image Enhancement. International Journal of Engineering Science and Innovative Technology (IJESIT) 1(2) (2012) 14. Abbas, Q., Farooq, A., Khan, M.T.A., Celebi, M.E., Garsia, I.F.: Feature Preserving Contrast Enhancement For Retinal Images. ICIC International 9(9), 3731–3739 (2013) ISSN 1349-1198 15. Feng, P., Pan, Y., Wei, B., Mi, D.: Enhancing: Retinal Image by the Contourlet Transform. Pattern Recognition Letters 28, 516–522 (2007)
Abstract. In medical image processing, retinal image enhancement is the challenging issue to reveal the unseen details of an retinal image, thus, in many applications image enhancement issued to solve the challenges such as, noise reduction, blurring, degradation, etc. To improve the visual grade of retinal images we have many alternative image enhancement techniques that are suitable for specific application. This paper presents an overview of various retinal image enhancement techniques that will process the original Retinal image to obtain enhanced image suitable for a specific application. The method used in this paper has been evaluated with help of PSNR image Quality measure which is applied over several retinal images which is obtained from the datasets such as DRIVE, STARE and few other’s provided by local medical experts. The comparative experimental results indicate that our proposed enhanced method has better outcome. Keywords: Image Processing, Retinal Images, Image analysis, Relative study, Retinal Image Enhancement Techniques, Contourlet Transform, Multi-scale decomposition.
1
Introduction
Image enhancement is used for the refinement of image quality, and to provide detailed information for Retinal image examiner, and also for any automated image processing techniques. During this refinement, one or more attributes of an image is reformed, for a particular task. Depending on the task the attributes are modified without spoiling original image. Spatial Domain method and Frequency Domain method are the two main categories in image enhancement. During the diagnosis and detection of many Retinal diseases, image enhancement techniques are used to recognize Diabetic Retinopathy (DR) and Age Related Muscular degradation (AMD). thus, refinement of an image is mandatory for diagnosis performed either automatically or manually A Retinal image shown in Fig.1, is affected by nonuniform illumination, in the captured image it can be seen that contrast and luminosity is not uniform. Thus the fundamental problem is to improve the quality of an image © Springer International Publishing Switzerland 2015 S.C. Satapathy et al. (eds.), Proc. of the 3rd Int. Conf. on Front. of Intell. Comput. (FICTA) 2014 – Vol. 2, Advances in Intelligent Systems and Computing 328, DOI: 10.1007/978-3-319-12012-6_64
581
582
P.S. Chandra, M.C. Hanumantharaju, and M.T. Gopalakrishna
during retinal image analysis. There are various filters that are available such as Gaussian filter, median filter, High pass filter and Low pass filter that are used to remove the noise that corrupt the retinal images. The Basic pre-processing steps for image enhancement are: 1. 2.
3.
Input image: - Image can be blur images, medical image etc. Pre-processing of an image: - Before applying various image enhancement techniques on the input images, various pre-processing methods are applied on those images. Applying domain techniques: - By using domain techniques quality of Pre-processed image is enhanced[12]. Such as, Log Transformation Technique Power Law Transformation Technique Alpha rooting Technique
Fig. 1. Retinal Image with Uneven Illumination and Contrast
2
Related Work
M Emre Celebi et al[14]. have proposed the feature preserving contrast enhancement Method for Retinal vascular Images using discrete-shearlet transform (DST) and perceptual uniform color space CIEL*a*b*. The study specify that Diabetic retinopathy and age related macular degeneration are the diseases that cause the retinal impairment and blindness. This method is widely used for both color and grayscale image enhancement, however this study does not consider the uneven illumination and the noise reduction during reconstruction of the image.
Retinal Based Image Enhancement Using Contourlet Transform
583
Peng Feng at al[15]. have proposed the concept of Contourlet transform over the retinal images using multi-scale edge enhancement. This study focus on the contrast enhancement which is necessary in pre-processing step for both natural images and ophthalmic images. In this method the image enhancement is performed using Contourlet transform which comprises of the following two steps: Laplacian pyramid (LP) and directional filter bank (DFB).
3
Proposed Method
Today, the Retinal Imaging is widely used in the diagnosis of retinal diseases such as blindness and visual impairments. The images obtained through digital devices contains uneven light illumination and some amount of noise, this results in poor quality of image which is not feasible for medical diagnosis. So far many methods are used world wide to enhance the retinal images, which have marginal drawbacks when compared. In this proposed method we improve the visibility of image optimally for the detection of various retinal diseases. For experimental purpose, various input retinal images have been obtained from the datasets DRIVE, STARE and few other’s provided by local medical experts. As indicated in Fig [2]. Initially the retinal image input obtained can be a gray scale image or color image. In an input image the edges may be brighter than the background area and surrounding edges may be weaker thus Discrete-shearlet Transform(DST) is applied on input retinal image. When the input is color image then Red, Green and Blue components are extracted. Since Green component has the optimal visibility we consider only green component as a primary source for the image enhancement. Now the green component is converted into gray-scale. Further CLAHE is applied on green component gray-scale image or any suitable HE method is applied as applicable to the type of image. After performing DST, at the time of image reconstruction there will be marginal uneven illumination and noise generated. Hence we use the methods such as Gamma correction which reduces the uneven illumination to obtain optimum image quality and suitable filters to remove different kinds of noise generated. Followed by Contourlet transform is applied. Finally we verify the image quality by using peak signal to noise ratio (PSNR) quality metric. By following the above proposed architecture, finally we get the optimal enhanced retinal image which has the best PSNR quality measure as experimentally compared.
584
P.S. Chandra, M.C. Hanumantharaju, and M.T. Gopalakrishna
Fig. 2. Proposed architecture
3.1
RGB Color Components
In any given color inputs, the RGB color components are extracted. By extracting these components we can observe the image clarities in all three components, in all the images of the extracted components, the green channel is observed to have significantly better visual quality than others as shown in Fig [3.1] below.
Fig. 3. Color components of RGB Image
Retinal Based Image Enhancement Using Contourlet Transform
585
The above selected green channel component is converted to gray-scale image, this gray-scale image can be feasible to apply gamma correction in reduction of uneven illumination, and noise reduction is possible by applying suitable filters. Thus we can infer that image is ready for applying contourlet transform. In an obtained input image the RGB color component extraction plays a vital role, further where the green channel component is responsible for image enhancement in getting the optimal result. 3.2
Contourlet Transform
The method Contourlet transform is associated with the decomposition of image into smaller sectors. By Contourlet transform we obtain sector wise smoother image along with smooth contours. The Contourlet transform technique overcome the challenges over traditional methods as wavelet transform and curvelet transform techniques. In this proposed method we make use of multi-scale decomposition. Multi-scale decomposition technique is best suited to get enhanced image. According to the study, Laplacian decomposition method is widely used for Image enhancement process. There exist a few drawbacks by using Laplacian pyramid in this method. Laplacian pyramid does not focus on oscillations in the images. Since the Laplacian Pyramid give chance of missing a few sub-bands during the reconstruction process image might not have the optimal resolution. Due to the above mentioned drawbacks, in the proposed method we make use of the multi-scale decomposition where the oscillations in the image is identified and we preserve the edges . By preserving edges the image obtained is smoother. The method of multi-scale decomposition focus on improving the contrast and preserve the contours quality. This makes the retinal image clearly visible and final outcome will have the optimal value. The scaling ratio is maintained uniform and during reconstruction process the quality of image is not lost, where as in traditional methods such as Laplacian pyramid this is a challenge to overcome.
(a)
(b)
Fig. 4. Contourlet Transform: (a) Input Image (b) Enhanced output Image
586
4
P.S. Chandra, M.C. Hanumantharaju, and M.T. Gopalakrishna
Experimental Results
For the experimental purpose we consider various retinal image inputs which are obtained from datasets such as DRIVE, STARE and few other’s provided by local medical experts. The obtained inputs are either gray scale or color image. The Experiment is carried out for all kind of images, and we could observe optimal results tested under various conditions. Below Fig [4.1] shows pictorial representation of the process where we can see the optimal enhanced image after applying contourlet transform, in which the obtained enhanced image is feasible for medical diagnosis. Compared to traditional methods, this proposed experimental result provides the highest PSNR value.
Fig. 5. Experimental Result: (a) Input color image (b) Extracted Green channel of color image (c) gray-scale of green channel image.
Fig. 6. Experimental Result: (a) Gamma correction after DST (b) CLAHE (c) Enhanced Image obtained using Contourlet Transform
5
Conclusion
The proposed method focuses on improving the visibility of image based on quality metrics like Edge Improvement Index, Contrast Improvement Index and Peak signal to Noise ratio. As evidenced by experiments with the Contourlet transform, there is a better preservation of contours than with other methods. Traditional methods does not consider the uneven illumination and noise reduction, hence the proposed method takes care of both to provide optimal image quality which is very appropriate for medical diagnosis. Finally proposed method tries to maintain a balance between PSNR, EII and CII, which will give us all these values in the accepted range and the comparative experimental results indicate that our proposed enhanced method has better outcome.
Retinal Based Image Enhancement Using Contourlet Transform
587
References 1. Jamal, I., Akram, M.U., Tariq, A.: Retinal Image Preprocessing: Background and Noise Segmentation. TELKOMNIKA 10(3), 537–544 (2012) 2. Bedi, S.S.: Various Image Enhancement Techniques- A Critical Review. International Journal of Advanced Research in Computer and Communication Engineering 2(3) (2013) 3. Joshi, G.D., Sivaswamy, J.: Colour Retinal Image Enhancement based on Domain Knowledge. In: Sixth Indian Conference on Computer Vision, Graphics and Image Processing (2008) 4. Shaeidi, A.: An Algorithm for Identification of Retinal Microaneurysms. Journal of the Serbian Society for Computational Mechanics 4(1) (2010) 5. Wang, Y., Tian, D., Vetro, A.: A Local Depth Image Enhancement Scheme for View Synthesis. Copyright Mitsubishi Electric Research Laboratories, Inc. (2012) 6. Maini, R., Aggarwal, H.: A Comprehensive Review of Image Enhancement Techniques. Journal of Computing 2(3) (2010) ISSN 2151-9617 7. Yahya, S.R.: Review on Image Enhancement Methods of Old Manuscript with Damaged Background. International Journal on Electrical Engineering and Informatics 2(1) (2010) 8. Vij, K., Singh, Y.: Enhancement of Images Using Histogram Processing Techniques. Int. J. Comp. Tech. Appl. 2(2), 309–313 (2009) 9. Kaur, N.: Image Enhancement Techniques a Selected Review. International Journal of Engineering Research and Technology (IJERT) 2(3) (March 2013) 10. Robinson, P.E., Lau, W.J.: Adaptive Multi-Scale Retinex algorithm for contrast enhancement of real world scenes (2012) 11. Kaur, E.M., Jain, E.K., Lather, E.V.: Study of Image Enhancement Techniques: A Review. International Journal of Advanced Research in Computer Science and Software Engineering 3(4) (2013) 12. Gavet, Y., Fernandes, M., Pinoli, J.-C.: Quantitative evaluation of image registration techniques in the case of retinal images. @JEIGavet 2012 (2012) 13. Mundhada, S.O., Shandilya, V.K.: Spatial and Transformation Domain Techniques for Image Enhancement. International Journal of Engineering Science and Innovative Technology (IJESIT) 1(2) (2012) 14. Abbas, Q., Farooq, A., Khan, M.T.A., Celebi, M.E., Garsia, I.F.: Feature Preserving Contrast Enhancement For Retinal Images. ICIC International 9(9), 3731–3739 (2013) ISSN 1349-1198 15. Feng, P., Pan, Y., Wei, B., Mi, D.: Enhancing: Retinal Image by the Contourlet Transform. Pattern Recognition Letters 28, 516–522 (2007)
