Image Enhancement for MR Images using LP-SVD

Technique

Bishnupriya Dash1, Debaraj Rana2 1.2Department of Electronics & Communication Engineering, Centurion University

1bishnupriyadsh233@gmail.com 2debarajrana85@gmail.com

Abstract— In this paper we have presented image enhancement technique which is based on Laplacian pyramid and Singular Value Decomposition for the improvement of visibility and better segmentation. Here we have used the shape invariant properties of Laplacian pyramid and singular value decomposition for the enhancement of sharpness of the image. This method not only sharpens the edges but also reduce the noise level after processing. The parameter PSNR and MSSIM has been evaluated after the processing for the performance evaluation. The proposed method has been implemented in MATLAB and the result show significance improvement in sharpening of MRI images.

Keywords— Image enhancement, Sharpening enhancement, Lapcian pyramid, Single value decomposition

I. INTRODUCTION

This Magnetic resonance imaging (MRI) is a technique that helps radiologists to identify pathology or structural abnormalities of subtle tissues. Despite, the clinical use of MR imaging modality, they are hampered by noise either due to imperfection in the RF coils or due to the problems associated with image acquisition. Due to an inadequate contrast between the miniature structure and its background, human vision strains accomplish better focus and frail identification of sharp edges is felt as an absence of region-of-interest (ROI). However, the visual quality of MR images is mainly focused on perceptual sharpness and precision of edges.

The objective of image sharpening technique is to emphasize fine details or augment the blurred edges. Therefore, sharpening concentrates on augmenting characters in the high-frequency domain. Image sharpening enhances the contrast by adding the scaled edge information near the boundaries of objects [1|. Sharpening is a crucial pre-processing approach to increase the contrast between the bright and the dark regions to bring out edge features. It enhances the subtle elements already present and improves the steepness of the edges. Since, the edges of subtle tissues/organs are of higher perceptual importance this research is focused on edge enhancement of subtle tissues.

Different image enhancement techniques have been proposed to enhance the visual quality and edge information in an image [2-5]. All these image enhancement techniques did image amplification, but these techniques have a drawback of producing undesirable artifacts at the boundary of the objects. Many research works have been presented on MR image enhancement for its contrast and identification of edge features. The most common MR image contrast enhancement methods are based on the morphological operation [6-8] and histogram equalization [9-11). All these methods are able to address the issue of contrast enhancement, but they do not perform well at the edges of subtle tissues/organs. Multi-scale methods have also been applied for the contrast enhancement of medical images. Two types of multi-scale methods are commonly adopted: Wavelet methods [12-16) and the LP methods [ 17.18,20].

This work presents a new sharpening enhancement method for MR images, using LP-SVD approach that can enhance the subtle tissue edges. Multi-scale images are decomposed into coarse and difference sub-bands by LP. Since the coarse sub-bands are inclusions of fine structures, the weighted sum of a singular matrix and its GHE of coarse sub-band images increase the contrast. The rest of the paper is organized as follows: Section 2 gives the materials and methods of the proposed method. Section 3 gives the experimental results and discussion of the proposed technique, and Section 4 gives the conclusion of the proposed method..

II. METHODS USED

The more pleasing visual quality of an image is obtained by emphasizing the edges. The degree of low sharpness around the object boundaries are explored in the spatial domain or missing high-frequency components in the frequency domain. Thus, the LPSVD sharpening enhancement technique certainly helps to sharpen the edges around the objects.

IAETSD JOURNAL FOR ADVANCED RESEARCH IN APPLIED SCIENCES

VOLUME VI, ISSUE V, MAY/2019

ISSN NO: 2394-8442

PAGE NO:29

A. Laplacian Pyramid (LP) The concept of LP on image decomposition was introduced by Burt and Adelson (21 ]. In LP decomposition, the multi-

scaled image is firstly low pass filtered using the Gaussian filter and down-sampled. The subsequent layer of LP (difference sub-bands) are determined by subtracting layer of the Gaussian pyramid (coarse sub-bands).

(1)

where denotes the down and up sampling of image by a factor of 2. is the filtered image of Xk and Lk is the successive layer of the LP.

B. Singular value decomposition (SVD) The SVD is a technique employed for data reduction, feature extraction and typically for enhancement of images. The SVD

decomposes the real and complex rectangular matrix A into a product of three matrices and is represented as (2)

where U and V are orthogonal matrices, and contains singular values of matrix A. The singular values of the matrix contain the intensity information. Therefore, any changes in this matrix lead to changes in the input image. Image equalization through SVD technique depends on adjusting the singular value matrix . The equalized SVD of A is written as:

(3)

where UA, VA are orthogonal square matrices called hanger and aligner respectively. matrix represents the sorted singular values on its main diagonal

III. PROPOSED LPSVD TECHNIQUE

Edges are an important characteristic of image since they correspond to object boundaries. An ideal edge is a scale invariant

in that no matter how much one increases the resolution the edge remains the same. LP consists of edge maps of the input image at different resolutions. Superimpose of LP (coarse sub-bands), and SVD techniques improve the sharpening by enhancing the contrast near object boundary, thus making the borders and edges visible better. The LP preserves the shape and phase of the edge maps across a scale. According to the proposed method, the edge map of the interested region is obtained through one level of LP. Then the edge details are sharpened by employing SVD techniques. One level of pyramid decomposition is employed in this proposed method because of loss of low frequency information at higher levels of the pyramid decomposition.

In the proposed LPSVD technique, the original image is decomposed into two different sub-bands C and D using one level of decomposition by LP as shown in Fig. 1(a). C is the coarse sub-band image of half size and D is the difference sub-band image of full size. Fig. 1(b) shows the proposed technique followed in coarse sub-band of LP. The low frequency coarse sub-band Cand histogram equalized coarse image Ce are scaled by SVD as following as:

(4)

(a)

IAETSD JOURNAL FOR ADVANCED RESEARCH IN APPLIED SCIENCES

VOLUME VI, ISSUE V, MAY/2019

ISSN NO: 2394-8442

PAGE NO:30

(b) Fig 1 (a) Flow diagram of proposed method

A. Evaluation Parameter

In this work, LPSVD enhancement technique has been compared with CHE, Wavelet. SWTSVD techniques. The subjective assessment of the proposed enhancement technique is evaluated by the following parameters:

(5)

(6)

(7)

IAETSD JOURNAL FOR ADVANCED RESEARCH IN APPLIED SCIENCES

VOLUME VI, ISSUE V, MAY/2019

ISSN NO: 2394-8442

PAGE NO:31

IV. RESULTS AND DISCUSSIONS

The visual comparison results of the proposed technique for various medical images are shown below. The LPSVD enhancement technique gives a better result when compared to CUE, Wavelet, and SWTSVD methods. The

undesirable artifacts of existing methods are overcome by LPSVD methods. The PSNR value of LPSVD method is high as it removes

The noisy artifacts from the MR images. Since, medical images contain more redundant information, the LPSVD based method retains it, whereas the GHE, Wavelet, and SWTSVD fail to preserve fine structures and details of the images. Table 2 shows the MSSIM values of various enhancement techniques. The values closer to 1 indicates better structural and edge preservation. Thus the LPSVD method is superior in edge preservation when compared to other methods. rior results when compared to other methods. The lower value of AMBE gives better brightness preservation of LPSVD based methods. The enhanced image ofGHE preserves brightness, but introduces artifacts at the edges. The enhanced image of Wavelet and SWTSVD fails to preserve small structural edge information and introduces blocking artifacts at their edges. The results are shown below:

(a) (b)

Fig 2 (a) Original input image (b) output of proposed method

(a) (b)

Fig 3 (a) Original input image (b) output of proposed method

IAETSD JOURNAL FOR ADVANCED RESEARCH IN APPLIED SCIENCES

VOLUME VI, ISSUE V, MAY/2019

ISSN NO: 2394-8442

PAGE NO:32

Here the orginal image has psnr_value=104.6436, MSE_value=2.2321e-06

Fig 4 Original image transform into wavelet domain

Fig 5.Original image into winner domain

(a) (b)

Org output

Original Image

Noisy Image Denoised Image1

IAETSD JOURNAL FOR ADVANCED RESEARCH IN APPLIED SCIENCES

VOLUME VI, ISSUE V, MAY/2019

ISSN NO: 2394-8442

PAGE NO:33

(c) (d)

Fig 6.Noisy Image and Denoised images

B. Pre-processing

In MR images, the presence of Rician noise lowers the contrast and degrade the visual quality of the image. Specifically, the

segmentation and measurement of CN size become a crucial task in computer-aided analysis of the image. Henceforth, pre-processing techniques such as de-noising and enhancement give promising results in segmentation and measurement of the CN. The visualization of the inner ear is proposed by Ergen et al. [26] for the investigation of benign paroxysmal positional vertigo. In this method, Otsu segmentation method has been utilized to segment the internal auditory canal and its nerves without pre-processing the MR image. In this proposed work the pre-processing is further improved by the sharpening enhancement technique.

The PSNR and MSSIM give the visual perception and the edge similarity of the pre-processed image. It is evident that the sharpening enhancement greatly influence the visual perception and edge preservation of the MR images. The segmentation results are varied based on the pre-processing techniques. The quantitative metrics shown in Table 4 depicts that the visual quality and edge preservation of enhancement technique is better when compared to de-noising technique. Therefore, de-noising and then sharpening enhancement methods have been employed as pre-processing for segmentation of the CN

V. CONCLUSION

This work presents an enhancement technique which not only enhances the visual quality, but also the segmenting capabilities of subtle organs from the MR images. The proposed method overcomes the problem of other enhancement techniques that produce the Gibbs phenomenon at edges. Being multi-scale, LP captures edges efficiently and these edge features are enhanced by SVD technique. However, the lines and contours of low contrast images introduce undesirable artifacts at the boundaries of subtle tissues. These artifacts can be reduced by the application of shearing operations of the algorithm in future work. The performance measures evaluated by parameter like PSNR, MSSIM.

REFERENCES

[1]. K. Konstantinides, V. Bhaskaran, G. Beretta, Image sharpening in the JPEG domain, IEEE Trans. Image Process. 8 (1999) 874-878. [2]. T.K. Kim, Contrast enhancement using brightness preserving bi-histogramequalization, IEEE Trans. Consum. Electron. 43 (1) (1997) 1-8. [3]. T.K. Kim, J.K. Paik, B.S. Kang, Contrast enhancement system using spatially adaptive histogram equalization with temporal filtering, IEEE Trans.

Consum. Electron. 44 (1) (1998) 82-86. [4]. T. Kim, H.S. Yang, A multidimensional histogram equalization by fitting an isotropic Gaussian mixture to a uniform distribution, Proc. IEEE Int.

Conf. Image Process., October 8 (2006) 2865-2868. [5]. H. Demirel, C. Ozcinar, G. Anbarjafari, Satellite image contrast enhancement using discrete wavelet transform and singular value decomposition,

IEEE Geosci. Remote Sens. Lett. 7 (2) (2010) 333-337. [6]. S. Satheesh, R.S. Kumar, K.V.S.V.R. Prasad, KJ. Reddy, Skull removal of noisy magnetic resonance brain images using contourlet transform and

morphological operations, in: 2011 International Conference on Computer Science and Network Technology (ICCSNT), IEEE, 4, 2011, pp. 2627-2631.

[7]. O.Gambino, E. Daidone, M. Sciortino, R. Pirrone, E. Ardizzone, Automatic skull stripping in MRI based on morphological filters and fuzzy c-means segmentation., in: 2011 Annual International Conference of the Engineering in Medicine and Biology Society, EMBC, IEEE, August, 2011, pp. 5040-5043.

[8]. C.C. Benson, V.L Lajish, Morphology based enhancement and skull stripping of MRI brain images., in: International Conference on Intelligent Computing Applications (ICICA), IEEE, 2014, pp. 254-257.

[9]. N. Senthilkumaran, J. Thimmiaraja, Histogram equalization for image enhancement using MRI brain images, in: World Congress on Computing and Communication Technologies (WCCCT), IEEE, February 2014, pp. 80-83.

Denoised Image2 Denoised Image4

IAETSD JOURNAL FOR ADVANCED RESEARCH IN APPLIED SCIENCES

VOLUME VI, ISSUE V, MAY/2019

ISSN NO: 2394-8442

PAGE NO:34

[10]. CM. Chen, C.C. Chen, M.C. Wu, G. Horng, H.C. Wu, S.H. Hsueh, H.Y. Ho, Automatic contrast enhancement of brain MR images using hierarchical correlation histogram analysis, J. Med. Biol. Eng. 35 (6) (2015) 724-734.

[11]. P.V. Oak, R.S. Kamathe, Contrast enhancement of brain MRI images using histogram based techniques, Int. J. Innov. Res. Electr. Electron. Instrum. Control Eng. 1 (3) (2013) 904.

[12]. J. Wu, X. Tian, Y. Sun, Z. Tang, A new wavelet-based adaptive algorithm for MR image enhancement, in: IEEE/ICME International Conference on Complex Medical Engineering, IEEE, 2007, pp. 600-603.

[13]. N.F. Ishak, M.J. Gangeh, R. Logeswaran, A preliminary study of high-field MRI image enhancement techniques applied to low-field MR brain images, in: 4th Kuala Lumpur International Conference on Biomedical Engineering, Springer Berlin Heidelberg, 2008, pp. 482-486.

[14]. Y. Yang, Z. Su, L Sun, Medical image enhancement algorithm based on wavelet transform, Electron. Lett. 46 (2) (2010) 120-121. [15]. D. Ray, D.D. Majumder, A. Das, Noise reduction and image enhancement of MRI using adaptive multiscale data condensation, in: 1st International

Conference on Recent Advances in Information Technology (RAIT), IEEE, March, 2012, pp. 107-113. [16]. P. Rasti, M. Daneshmand, F. Alisinanoglu, C Ozcinar, G. Anbarjafari, Medical image illumination enhancement and sharpening by using stationary

wavelet transform, in: Signal Processing and Communication Application Conference (SIU), IEEE, 24, 2016, pp. 153-156. [17]. D. Frejlichowski, R. Wanat, Application of the Laplacian pyramid decomposition to the enhancement of digital dental radiographic images for the

automatic person identification, Image Anal. Recognit. (2010) 151-160. [18]. S. Dippel, M. Stahl, R. Wiemker, T. Blaffert, Multiscale contrast enhancement for radiographies: Laplacian pyramid versus fast wavelet transform,

IEEE Trans. Med. Imaging 21 (4) (2002) 343-353. [19]. M.Z. Iqbal, A. Ghafoor, A.M. Siddiqui, M.M. Riaz, U. Khalid, Dual-tree complex wavelet transform and SVD based medical image resolution

enhancement, Signal Process. 105 (2014) 430-437. [20]. H. Greenspan, C.H. Anderson, S. Akber, Image enhancement by nonlinear extrapolation in frequency space, IEEE Trans. Image Process. 9 (6) (2000)

1035-1048. [21]. P. Burt, E. Adelson, The Laplacian pyramid as a compact image code, IEEE Trans. Commun. 31 (4) (1983) 532-540. [22]. S. Murty, P.R. Kumar, A robust digital image watermarking scheme using hybrid DWT-DCT-SVD technique, Int. J. Comput. Sci. Netw. Secur. 10

(1) (2010)185-192.

IAETSD JOURNAL FOR ADVANCED RESEARCH IN APPLIED SCIENCES

VOLUME VI, ISSUE V, MAY/2019

ISSN NO: 2394-8442

PAGE NO:35