Salient Features Extraction in Medical Images Using Sparse Contourlet

Content-based Image Retrieval Using the Dual-Tree Complex Wavelet Transform Saliency filters: contrast based filtering for salient region detection
Dr.ShawnPitt Profile Pic
Dr.ShawnPitt,Netherlands,Teacher
Published Date:17-11-2017
Your Website URL(Optional)
Comment
I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 Published Online September 2017 in MECS (http://www.mecs-press.org/) DOI: 10.5815/ijigsp.2017.09.01 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation 1 2 Rami Zewail , Ahmed Hag-ElSafi University of Alberta, Edmonton, Alberta, Canada Smart Empower Innovation Labs Inc., Edmonton, Alberta, Canada 1 2 Email: rami.zewailsmartempower.ca , ahmed.hagelsafismartempower.ca Received: 12 May 2017; Accepted: 05 July 2017; Published: 08 September 2017 Abstract—Medical experts often examine hundreds of x- the last decade, concepts of sparsity of representation and ray images searching for salient features that are used to its applications in computer vision has been gaining an detect pathological abnormalities. Inspired by our increased interest from the research community. In the understanding of the human visual system, automated medical imaging paradigm, sparsity techniques have been salient features detection methods have drawn much used in applications such as image enhancement, attention in the medical imaging research community. segmentation, quantification of diseases. In this paper, we However, despite the efforts, detecting robust and stable present a novel appearance-based salient feature salient features in medical images continues to constitute extraction and matching method based upon a sparse a challenging task. This is mainly attributed to the Contourlet-based representation. The multi-scale and complexity of the anatomical structures of interest which directional capabilities of the Contourlets is utilized to usually undergo a wide range of rigid and non-rigid extract salient points that are robust to noise, rigid and variations. non-rigid deformations. Moreover, we also include prior In this paper, we present a novel appearance-based knowledge about local appearance of the salient points of salient feature extraction and matching method based on the object of interest. This allows for detection a reduced sparse Contourlet-based representation. The multi-scale set of robust and stable salient points that is most relevant and directional capabilities of the Contourlets is utilized to the structure of interest. to extract salient points that are robust to noise, rigid and The rest of the paper is organized as follows: Section II non-rigid deformations. Moreover, we also include prior covers some of the related work in literature. Section III knowledge about local appearance of the salient points of presents the details of the proposed Appearance-based the structure of interest. This allows for extraction of salient feature extraction method. Experiments and results robust stable salient points that are most relevant to the are presented in section IV. Finally, conclusions are anatomical structure of interest. drawn in section V. Index Terms—Salient features, multiscale, appearance, sparse, contourlet II. RELATED WORK Inspired by the Human Visual System, salient (interest) point detection methods have been used in many I. INTRODUCTION computer vision applications such as registration, Within the field of medical imaging, salient feature tracking, robot navigation, and recently in medical extraction has drawn much attention in a wide range of imaging. Salient points need to be well defined and stable applications. Examples include: computer-aided under various transformations. Various salient point diagnosis, segmentation, registration, and image retrieval. detection algorithms are often evaluated in terms of its However, detecting robust and stable salient features in Repeatability; that is the ability of the method to detect medical images continues to constitute a challenging task. the same salient points under various transformations This is mainly attributed to the high resolution of medical such as: scaling, rotation, and change in view. images, and complexity of the anatomical structures of Several methods have been proposed in the literature interest which usually undergo a wide range of rigid and for salient point detection. In 1, Schmidt et. al. non-rigid variations. classified salient detection methods into three classes: In the era of Big Data and mobile-based healthcare, edge-based, intensity-based, and biologically-inspired these challenges have even escalated where there is a methods. The Harris corned detector, 2, is among the growing demand for analyzing of high resolution medical most famous methods in the literature. It detects corners images and large scale biomedical data in general. Over in an image based upon the autocorrelation matrix. Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 2 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation Despite being useful in simple images, the Harris it is beneficial for the salient point detector to be detector suffers from the following limitations: it is based target driven. Thus, it is desirable to integrate top- on single scale analysis, and has poor localization ability. level prior knowledge within the salient feature Another famous salient point detection method is the extraction process. SIFT algorithm in 3. The SIFT algorithm extracts salient points by finding the local maxima of the response The method is based upon Non Sub-sampled of the Difference-of-Gaussian (DOG) operator. Contourlet Transform (NSCT). The multi-scale and Biologically-inspired salient point detectors have focused directional capabilities of the NSCT transform is utilized onto directional and multi-scale analysis of the images. Q. to extract salient points that are robust to noise, rigid and Tian 4 and Loupais et. al. 5 proposed wavelet-based nonrigid deformations. We also include prior knowledge salient point detectors. The wavelet transform has several about the local appearance of the salient points of the appealing properties such as the multi-scale pathology of interest. This enables us to extract robust representation and the spatial-frequency localization. and stable salient points that are most relevant to the Nevertheless, wavelets do suffer from lack of shift pathology of interest. invariance and limited orientation selectivity. These are First, salient features are extracted in a bottom-up both important properties for efficient detection of salient approach using Non-subsampled Contourlet-based points in medical images with complex anatomical representation. Next, local appearance profile for the structures. In 6, itti and Koch proposed a biologically pathology of interest is built using Guassian Mixture inspired bottom-up saliency detector based upon Gabor Model (GMM). The details of these two steps are filters. In 7, the authors presented a salient point presented in sections III.A and III.B consequently. detector based upon the phase congruency of Log-Gabor A. Contourlet-based Bottom-up Salient Feature filter response. In 8, Xinting Gao et. al. proposed a Extraction multi-scale corner detection method based upon Log- Gabor transform. The bottom-up multi-scale salient feature extraction In 11, Perazzi et al. used contrast-based filtering for step is achieved using Non-subsampled Contourlet detection of salient regions. In 12, the authors used representation(NSCT) and second moment matrix. Using Harris method to build a saliency map that is used in the ability of the NSCT transform to capture higher order segmentation. In 13, Dual tree complex wavelet image structures, we are able to extract robust salient transform is used in a content-based image retrieval features in the medical images. We will refer to the new application. In 14, the authors used a deep network method as: NSCT-SMM. Fig.1 shows a block diagram architecture to build a linear invariant feature transform for the proposed method. (LIFT). In 15, Mark Brow et al. presented a generalized framework for salience feature detection using histogram. In 17, the authors proposed a method for salient point detection based on steerable filters and Harris point detector. In 18, Omprakash et. al. proposed a method for fast multi-scale visual saliency detection using scale space interpolation. III. METHODOLOGY We next present the details of the proposed salient feature extraction and matching method. The new method is suited for extraction of robust salient features of interest in medical images. The proposed method attempts to incorporate the following characteristics:  Biologically inspired: Since there no strict definition of what constitutes a salient point in an image, it is logical to use biologically-inspired filters in the detection process.  Localized: The localization property of the detector is also an important aspect that tends to affect other higher-level tasks such as feature extraction, Fig.1. Flowchart of the proposed Contourlet-based bottom-up salient feature extraction step (NSCT-SMM) matching, and registration.  Able to detect higher order salient structures in A multi-scale directional decomposition is achieved images (e.g. contours). using the Non Sub-sampled Contourlet transform. The  Incorporate prior knowledge about the pathology of input image I(x, y) is decomposed into 4 scales. In the interest: In most medical image analysis applications, we are interested in a structure; hence first scale level, we use a 4-level Directional Filter Bank Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation 3 N (DFB) to get 16 directional frequency partitioning. In the j M 2 second and third scales, eight directional frequency cW ( sin( )) (3)  j,k k jk  00 partitions are obtained using a 3-level Directional Filter bank. In the final scale level, 2-level directional filter- Where j is the scale of the decomposition, M is the banks are used. Let Bj , 0,1,2,3 denotes the set of  j k maximum number of scales, is the orientation at response images that results from the non sub-sampled scale j , is the number of decomposition orientations pyramid decomposition step, where j is the scale level N j At each scale j , the response image B is further at scale j , is the magnitude of the non-subsampled W j jk , decomposed using l -level non sub-sampled Directional k j contourlet coefficients at scale j and orientation , and th Filter Bank to produce a set of multi-scale multi- (k1) is the angle of the k orientation at scale  jk , N directional contourlet sub-band images W , where  j jk , l1 level j . j k 0,1,...,2 , and . j 0,1,2,3 The second moment matrix is then constructed as: Fig.2 shows the oriented contourlet basis functions for two different scales. Due to the flexibility of the ab transform, we can decompose the image using different  M (4) angular resolutions at every scale.  (4) bc  The eigenvalues of the second moment matrix in equation 4 is then calculated as: 11 22 (5) E a c 4b (a c) (5) 1 22 (6) 11 22 E a c 4b (a c) (6) 2 22 Next, we define a contourlet-based feature strength image, F , as a weighted sum of the two eigenvalues of the second moment matrix. The smaller eigenvalue, E , is 1 a measure of the corners in the image and the larger eigenvalue, E , is a measure of the contours and edges in 2 the image 9. Fig.3 shows an example of x-ray image of human spine and the corresponding Contourlet Feature strength map. F E (1 )E (7) (7) 12 Salient features are finally extracted from the Contourlet-based feature image through non-maximum suppression and thresholding. This step selects local maxima points in the Feature Image as salient points. Fig.2. NSCT basis functions for two scales with different angular B. Incorporation of prior pathology-related knowledge resolutions at each scale/(a) 16 directional basis functions at scale1, (b) 8 directional basis functions at scale 2. Finally, we incorporate prior knowledge about the structure of interest to bias the bottom-up salient point Next, following classical moment analysis equations, detection procedure detailed in III.A. This allows us to 9,16, we compute the following: extract a more stable and robust set of salient features that are most relevant to the pathology of interest. The new N M j 2 method benefits from the rich multi-scale directional aW ( cos( )) (1)  j,k k information offered by the Non-Subsampled Contourlet jk  00 decomposition to construct localized descriptors for N salient points of the structure of interest. The local M j (2) bW cos( )sin( ) descriptors are then modeled using Mixture of Gaussians  j,k k k jk  00 Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 4 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation to construct a local appearance model of the target salient sub-bands at every scale, excluding the low-pass response. points. The average absolute deviation of the gray values of the square grid for each sub-band is calculated as follows: NN a lb 22 1 i i i (8) E (a,b) x (l,k) (8) j  j j 2 N NN la lb 22 th Where suppose that x denotes the j NSCT-based j i Feature image. is the NSCT coefficient of the x (l,k) j i x NSCT Feature image located in (lk , ) . is the  j j mean of the coefficients in the NxN square grid whose center is the characteristic salient point P . The i contourlet-based local descriptor for salient point P is i then defined as: T F E , E , E ,...., E (9) iD 1 2 3 (9) Where E is the average absolute deviation in a square i th grid for the NSCT sub-band. D is the total number of i NSCT sub-bands. Fig.3. Example x-ray image and corresponding Contourlet-based Feature Strength Image. (a) original x-ray image, (b) contourlet-based feature strength image, (c) 3D view of the contourlet-based feature image. Given a set of M training images with a ground truth segmentation for the structure of interest, we extract a set of N ground truth salient points. First, we use the method described in section III.A to extract salient features in the image. Next, the ground truth salient points that correspond to the structure of interest are selected using the ground truth segmentation. A ground truth salient point is defined as: a salient point that has a point-to- curve distance less than 5 pixels. Fig.4 shows two x-ray images with the ground truth salient points that define the border of the lumbar vertebrae L1-L5. For every ground truth, salient point ( P ), we extract a i local descriptor based upon Contourlet-based representation. We use a square grid of NxN (N=7) points around each ground truth salient point. Variance-based Fig.4. Example of ground truth salient points that defines the human features are then extracted from the NSCT directional lumbar spine in x-ray images. Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation 5 11 (11) T1 f (F ) exp (F ) ( ) (F ) n n n D 0.5 n (2) ( ) 2  k N m  : Mixture weights, where  1 n  k n1 The parameters of the Gaussian mixture model are as follow: θ  , , ,nN 1.. (12)  n n m n Using the training set F , we estimate the  k kK1.. parameters of the Gaussian mixture model (GMM) by maximizing the likelihood function. The GMM likelihood function is given by: K p(F /θθ )  p(F / ) (13) k k1 k kK1.. We use the Expectation-Maximization algorithm (EM- algorithm) to estimate the model parameters that maximize the likelihood function given in equation 13. θθ arg max pF ( / ) (14) θ k kK1.. Fig.5. Extraction of contourlet-based local descriptor for each ground The prior local appearance model is then used to bias truth salient point the outcome of the bottom-up salient feature extraction The outcome of this stage is a set of K=NxM step described in section III.A. For every detected low- contourlet-based feature vectors extracted from the level salient point, we extract a contourlet-based local ground truth salient points in the training set. Fig.5 descriptor in the same manner as explained in Fig.5. Next, summarizes the procedure for construction of the we estimate the confidence level that the detected salient contourlet-based local descriptor. point belongs to the model learnt during the training We then use the matrix of local descriptors to construct phase. To do so, following Bayes decision rule 10, we a local appearance model for the structure of interest use the posterior probability, as a measure for the using Mixture of Gaussians. confidence level. Let F be the set of contourlet-based local  i iK1.. descriptors for the ground truth salient points in the IV. EXPERIMENTS AND RESULTS training set. K is the total number of ground truth salient th points, and F is the local descriptor of the i salient We evaluate the performance of the proposed salient i feature extraction method using various test images. Fig.6 point. We model the local descriptors of the ground truth shows examples of the images used in the experiments. salient points in the training set as a mixture of N m Fig.6(a) and Fig.6(b) are commonly used in the literature Gaussians. The Gaussian Mixture model is a flexible to evaluate salient point detection algorithms. Fig.6(b) is parametric distribution 10. an example of the x-ray images of the human spine in the The Gaussian mixture model for the local descriptors is evaluation process. The synthetic image in Fig.6(a) given by: exhibits different boundaries with varying strength levels. It is used to evaluate the localization capabilities of the N m (10) new method. The “Van Gogh” image shown in Fig.6(b) is p(F /θ) f (F ) k n n k n1 a famous painting that contains a lot of texture. It is used to evaluate the robustness of the proposed method against Where N : Number of Gaussian functions in the various geometric and photographic transformations; m these include: rotation, illumination, and noise. Fig.6(c) model. shows example of x-ray image of human spine that were fF() : Multivariate Gaussian density functions nk used to test the performance of the proposed method characterized by the mean ( ) and covariance matrix n when applied to medical images. ( ).  n Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 6 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation is able to reveal most of the salient points in the image. These include corners, strong and weak edges, and smooth contours. On the other side, the Harris detector and the Laplacian-of-Gaussian (LOG) detector seem to perform good only in case of strong salient locations (corners and edges), while both detectors show limited localization capabilities in case of weak edges and smooth contours. The phase congruency method (PC- based) demonstrates better performance compared to the Harris and LOG detectors. However, the proposed NSCT-based detector appears to be superior in localizing Fig.6. Images used for evaluation of NSCT-based salient point detector. salient points along the smooth circular contour. This is (a) Synthetic image, (b) VanGogh Image, (c) X-ray image of human attributed to the fact that Non-subsampled Contourlet spine Transform (NSCT) offers more flexible directional analysis compared to Log-Gabor filters. Similar We evaluate the robustness of the proposed salient observations are obtained when the salient detectors were point detector using the Repeatability Rate criterion ( rr ) applied to the x-ray image of human spine in Fig.8. introduced by Schmid et. al. 1. Given two images that Among all detectors investigated, the proposed detector are related together by some transformation, the seems to better detect anatomical structure with poor repeatability rate is the ratio between the number of edges and smooth contours. corresponding points that are detected in both images to the total number of points in the common area. In our experiments, we consider the following relationships between the reference and test images:  Rotation transformation.  Thin Plate Spline transformation (as an example of non-rigid transformation).  Variations in illumination levels.  Variations in noise levels. (rr ) ( ) (tt ) ( ) Let  X , and be two sets  X i j of interest points detected in the reference and test images respectively. Let be the number of salient points in N c (rt ) ( ) the common area. Let be the number of CX , X ij corresponding pairs of salient points; with a localization rr error of 1.5 pixels. Then, the Repeatability rate ( ) between the two images is defined by: Fig.7. Localization of various salient point detectors. (a) Harris detector, (rt ) ( ) (b) Laplacian detector, (c) Phase Congruency detector (PC-based), and CX , X ij (15) rr (d) NSCT-based detector. N c We evaluate the proposed method (NSCT-based) against the following famous salient point detectors from the literature: 1. Harris detector (Harris). 2. The detector proposed by Kovesi et. al. in 9 that is based upon Log-Gabor decomposition and phase congruency (PC-based). 3. A salient point detector that is based upon steerable pyramid decomposition instead of Non sub-sampled contourlet decomposition (Steerable- based). A. Localization The localization capabilities of the proposed method Fig.8.Localization of various salient point detectors in medical images. are evaluated using the synthetic test image shown in (a) proposed NSCT-based detector, (b) Harris detector, and (c) Phase Fig.6 (a). Fig.7 shows salient points extracted using Congruency detector. various salient point detectors. The NSCT-based detector Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation 7 achieved for noise variance levels below 0.02. At higher B. Robustness to rotation noise levels, the proposed method shows much more robustness to increased noise levels with an average Image rotations are obtained by rotating the image around its central axis. Test images for evaluation of the improvement in the repeatability rate of 70.40 %. interest point detectors can be found in 10. We use a set of rotated VanGogh images. The repeatability rate is calculated as a function of the rotating angle 10-180. Fig.9 shows the performance of the proposed salient point detector compared to several famous salient point detectors including Harris detector, Cottier detector, and Heitger detector 1. The Harris detector and Cottier detector are based upon auto-correlation matrix. Heitger detector is based upon Gabor-like filters. The proposed NSCT-based detector appears to outperform all three detectors at various rotation angles. Fig.11. Effect of variations in noise levels on the repeatability rate for various salient point detectors D. Robustness to change in illumination The robustness of the proposed method to uniform change in illumination is evaluated using the VanGogh test image with changing levels of brightness. This is  done by changing the illumination factor ( ). Fig.12 shows how the repeatability rate is affected by altering Fig.9. Performance of the proposed salient feature extraction method under various rotations (a) VanGogh test image at different rotations, (b) the brightness level of an image a positive value of  Corresponding salient features extracted at each rotation position. increases the brightness level, and a negative value of  decreases the brightness level. Compared to Harris detector, the proposed NSCT-based detector has achieved an average improvement in repeatability rate of 48.89%. Compared to the Phase congruency–based detector (PC- based), 9, our method achieves an improvement of 13.124 % in the repeatability rate. Fig.10. Robustness to rotation of various salient point detectors C. Robustness to noise The robustness to noise is evaluated using the VanGogh test image with added Gaussian noise. Fig.11 shows the effect of the increased noise level on the Fig.12. Effect of brightness level on the repeatability rate for various repeatability rate of the salient points for both the salient point detectors proposed NSCT-based metho, and Harris detector. From Fig.11, we observe the following: an average improvement in the repeatability rate of 15.23 % is Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 8 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation Fig.14 demonstrates the benefit of incorporating prior E. Robustness to non-rigid transformations pathology-related knowledge with the bottom-up salient feature extraction process. By applying the pathology- In medical images, anatomical structures undergo a range of non-rigid deformations. Hence, a good salient related prior knowledge, an average reduction ratio of 51.86% is achieved in the extracted salient features. This point detector for medical images needs to be robust to non-rigid deformations in the object of interest. We next allows for extraction of features that are robust and most relevant to the pathology of interest. evaluate the robustness of the new method to non-rigid deformations. Experiments are conducted using a set of x-ray images of lumbar spine with different non-rigid deformations. The structure of interest in every image pair is related by a non-rigid transformation estimated using the Thin Plate Spline model. For each method, we calculate the repeatability rate as the percentage of correct matching to the total number of common points between the two images. Table 1 compares the achieved average repeatability rate for the proposed NSCT-based detector, Harris detector, PC-based detector, and the steerable-based detector. The proposed NSCT-based detector achieves improvements of 64.224 %, 18.0 % in the average repeatability rate compared to Harris detector, and PC- based detector respectively. The NSCT-based detector achieves an improvement of 9.95% in the average repeatability rate compared to the Steerable-based Fig.14. Example for salient point detection using NSCT-based method detector. (a) with no top-level knowledge, (b) with top level appearance knowledge. G. Performance in medical image retrieval applications Finally, we evaluate the suitability of the proposed appearance-based salient feature extraction method in medical image retrieval applications. We use images from the IRMA database which is made publicly available as a part of the Image Retrieval for Medical Applications (IRMA) project. The database consists of radiograph images that were randomly collected from regular routines at the department of diagnostic Radiology, Aachen University of Technology (RWTH), Aechen, Germany. In our experiments, we use a subset of 1000 images, where 800 images used in training and 200 images left out for testing as query images. We evaluate the performance of the new method using the following criteria: - Average retrieval rate (ARR): For a query image, the retrieval rate is the ratio between the number of relevant Fig.13. Example of two x-ray images where the object of interest is images retrieved to the total number of retrieved images. related by non-rigid deformations ARR The Average retrieval rate, , is calculated by averaging the retrieval rates over all query images. Table 1. Evaluation of robustness of different salient point detectors - Average Error Rate (ER): This is the percentage that to non-rigid deformations of the object of interest the first retrieved image is not a relevant image averaged over all query images. Salient Point detector Average repeatability rate method Proposed method NSCT- 0.7891 Fig.15 evaluates the performance of the proposed based system using average retrieval rates (ARR) as a function Steerable-based 0.7177 of the number of top retrieved images. We compare the Harris detector 0.4805 performance of the following systems: PC-based detector 9 0.6687 i. The proposed retrieval system using Contourlet- F. Appearance-based prior knowledge based salient point detectors (NSCT-local). Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation 9 ii. Retrieval system based upon Gabor-based salient sub-bands. The new method also integrates prior local detectors (Gabor-local). appearance knowledge of the structure of interest in the iii. Retrieval system based upon global features detection process. Compared to related methods in the extracted from the response of the NSCT sub- literature, our method has proven successful in detecting bands (NSCT-global). stable and relevant salient points in medical images. The proposed method opens door for pathology-related The proposed retrieval method (NSCT-local) tends to machine learning tasks such as image retrieval, consistently outperform the other two methods. Over a automated detection of abnormalities and diseases. range of 5-40 top retrieved images, the proposed method achieves an average improvement in the Average ACKNOWLEDGEMENT Retrieval Rate (ARR) of 9.574 % and 17.338 % over the The authors wish to acknowledge the US. National Gabor-based system, and the NSCT-global system Library of Medicine for making the digital x-ray images respectively. available. REFERENCES 1 C. Schmid, R. Mohr and C. Bauckhage, “Evaluation of Interest Point Detectors,” International Journal of Computer Vision, vol. 37(2), pp.151-172, 2000. 2 C. Harris, M. Stephens, “A combined corner and edge detector,” Proceedings of Alvey Vision Conference, 1988. 3 D. Lowe, “Object Recognition from Local Scale-Invariant th Features,” Proceedings of 7 IEEE International Conference on Computer Vision, vol. 2, pp. 1150-1157, 1999. 4 Q. Tian, N. Sebe, M. S. Lew, E. Loupias, and T. S. Huang, “Image retrieval using wavelet-based salient points,” Journal of Electronic Imaging, vol. 10(4), pp. 835-849, 2001. 5 E. Loupias, N. Sebe, “Wavelet-based Salient Points: Applications to Image Retrieval Using Color and Texture Fig.15. Average retrieval rates versus number of retrieved images for Features,” Proceedings of 4th Int. Conf. on Visual the following image retrieval methods: Gabor-based salient points Information Systems, pp. 223-232, 2000. (Gabor-local), NSCT-based salient points (NSCT-local), and NSCT- 6 L. Itti, C. Koch and E. Niebur, “A model of saliency- based global response (NSCT-global). based visual attention for rapid scene analysis,” IEEE Transaction on Pattern Analysis and Machine Intelligence, Table 2 evaluates the performance of the proposed vol. 20(11), pp. 1254-1259, 1998. method in terms of the average error rate (AERR). The 7 C. Wu, Q. Wang, “A Novel Approach for Interest Point proposed method (NSCT-local) achieves an average error Detection Based on Phase Congruency,” Proceedings of rate of 21.43 % versus 35.71 % and 46.43 % for the TENCON 2005, pp. 1-6, 2005. 8 X. Gao, F. Sattar, R. Venkateswarlu, “Multiscale Corner Gabor-based system, and the NSCT-global system Detection of Gray Level Images Based on Log-Gabor respectively. Wavelet Transform,” IEEE Transactions on Circuits Systems and Video Technology, vol. 17(7), pp. 868-875, Table 2 Average error rates for different content-based image 2007. retrieval methods 9 P. Kovesi, “Phase Congruency Detects Corners and Method Average Error Rate Edges,” Australian Pattern Recognition Society NSCT-local 21.43 % Conference: DICTA 2003, pp. 309-318, 2003. nd 10 A. Webb, Statistical Pattern Recognition. 2 edition, 2002. NSCT-global 46.43 % 11 Perazzi, F., Krahenbuhl, P., Pritch, Y., Hornung, A.: Gabor-local 35.71 Saliency filters: contrast based filtering for salient region detection. In: 2012 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pp. 733–740. IEEE (2012). V. CONCLUSION 12 Hangbing Gao, Yunyang Yan, Youdong Zhang, Jingbo In this paper, we presented a novel appearance-based Zhou, Suqun Cao, and Jianxun Xue “Automatic salient feature extraction that can extract robust and Segmentation of Nature Object Using Salient Edge Points Based Active Contour,” Mathematical Problems in stable salient features in medical images. The salient Engineering, Volume 2015 (2015), Article ID 174709. feature extraction process is performed in a sparse 13 Stella Vetova, Ivan Ivanov, Content-based Image contourlet-based domain. The proposed method is based Retrieval Using the Dual-Tree Complex Wavelet upon Non Sub-sampled Contourlet Transform (NSCT), a Transform,” Proceedings of MCSI '14 Proceedings of the truly 2D multi-scale and multidirectional image analysis 2014 International Conference on Mathematics and tool. Rich discriminative information was extracted using Computers in Sciences and in Industry, pp. 165-170. local descriptors from the non-subsampled contourlet Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10 10 Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation 14 K. M. Yi, E. Trulls, V. Lepetit and P. Fua, “LIFT: learning and predictive analysis, and medical image processing. Learned Invariant Feature Transform,” European He is currently the Staff Researcher-Machine Learning at Smart Conference on Computer Vision (ECCV) 2016. Empower innovation labs Inc., Alberta, Canada. 15 Mark Brown, David Windridge, Jean-Yves Guillemaut, Mr. Hag-ElSafi is a member the Smart City Alliance in “A generalized framework for saliency-based point Alberta, Canada and the Association of Professional Engineers feature detection,” Computer Vision and Image and Geoscientists of Alberta (APEGA). Understanding, Volume 157, April 2017, pp. 117–137. 16 M. Nixon, A, Aguado. Feature Extraction and Image nd Processing. 2 edition, Elsevier, 2008. Rami Zewail received a BSc. and MSc. 17 Mahesh, Subramanyam M. V,"Feature Based Image In Electronics & Communications Mosaic Using Steerable Filters and Harris Corner Engineering, Arab Academy for Science Detector", IJIGSP, vol.5, no.6, pp.9-15, 2013.DOI: and Technology, Egypt, in 2002 and 2004 10.5815/ijigsp.2013.06.02. respectively. And a PhD degree in 18 Omprakash S. Rajankar, Uttam D. Kolekar, “Scale Space Electrical and Computer Engineering, Reduction with Interpolation to Speed up Visual Saliency University of Alberta, Canada, in 2010. Detection", IJIGSP, vol.7, no.8, pp.58-65, 2015.DOI: He has over 15 years of academic and 10.5815/ijigsp.2015.08.07 industrial R&D experience in areas of machine learning, digital signal processing, and embedded computing. He has contributed to the scientific community with over 15 publications in areas of image processing, statistical modeling, and embedded computing. Currently, he is the Authors’ Profiles Principal Researcher for Machine Learning at Smart Empower Innovations Labs Inc., Edmonton, Alberta, Canada. Ahmed S. Hag-ElSafi (Khartoum,1978) Dr. Zewail is a member of the Institute of Electrical and holds a BSc and a MSc of Electronics and Electronics Engineers (IEEE) and the Association of Communications Engineering, Arab Professional Engineers & Geoscientists (APEGA). He served as Academy for Science and Technology, a reviewer for the Journal of Electronics Imaging, and Journal 2002 and 2004 respectively. of Optical Engineering for the SPIE society, USA. He has 15 years of experience of research and industrial development in the areas of embedded systems and machine learning. He has published more than fifteen papers in the areas of IOT security, biometrics, machine How to cite this paper: Rami Zewail, Ahmed Hag-ElSafi,"Appearance-based Salient Features Extraction in Medical Images Using Sparse Contourlet-based Representation", International Journal of Image, Graphics and Signal Processing(IJIGSP), Vol.9, No.9, pp.1-10, 2017.DOI: 10.5815/ijigsp.2017.09.01 Copyright © 2017 MECS I.J. Image, Graphics and Signal Processing, 2017, 9, 1-10