A New Feature-Based Image Registration Algorithm

D Computer Technology and Application 4 (2013) 79-84 DAVID PUBLISHING A New Feature-Based Image Registration Algorithm Md. Baharul Islam1 and Mir ...
Author: Sabina Page
6 downloads 1 Views 331KB Size
D

Computer Technology and Application 4 (2013) 79-84

DAVID

PUBLISHING

A New Feature-Based Image Registration Algorithm Md. Baharul Islam1 and Mir Md. Jahangir Kabir2 1. Department of Multimedia Technology and Creative Arts, Faculty of Science and Information Technology, Daffodil International University, Dhaka 1207, Bangladesh 2. Department of Computer Science and Engineering, Faculty of Electrical and Computer Engineering, Rajshahi University of Engineering & Technology, Rajshahi 6204, Bangladesh

Received: February 6, 2013 / Accepted: February 18, 2013 / Published: February 25, 2013. Abstract: IR (Image Registration) is one of the important operation of image processing system which is the process of aligning two or more images into one coordinate system that are taken at different times, from different sensors, or from different viewpoints. It has a lot of applications especially medical imaging and remote sensing. The main purpose of this paper is to provide a comprehensive review of existing literatures available on image registration system and proposed a new feature-based IR technique using edge of images. We used edges as a feature of images for registration. It will be a useful document for researchers who will work on feature-based image registration regardless for specific applications. Key words: Feature detection, image registration, feature extraction, transformation.

1. Introduction The resulting images need to geometrically align to one another to get better observation. There are a lot of applications of IR (Image Registration) such as motion correction, correcting of geometric distortion, formulation of composite functional maps, mapping of PET (Positron Emission Tomography) or SPECT (Single Photon Emission Computed Tomography) to MRI (Magnetic Resonance Imaging) and many more. Reference [1] are given the details applications of IR in different fields. There are a lot of image features that can be used for feature-based image registration. I preferred edge as a feature of an image that covered surrounding of images. Edge detection is divided by two categories gradient and Laplacian. The gradient method detects the edges by finding for the maximum and minimum in the first derivative of the image whereas the Laplacian method searches for zero

Corresponding author: Md. Baharul Islam, senior lecturer, M.Sc., research fields: image processing, medical imaging, modeling and simulation, animation, digital media. E-mail: [email protected].

crossings in the second derivative of the image. I took gradient method for edge detection in my experiment because of two filters like horizontal and vertical. Basically there are four steps involving in image registration system. It involves finding salient features in the two or more images to be registered. It may be point detection, edge detection, corner detection, pixels, textures, colors, histogram etc. The resulting features will be subsets of the image domain, often in the form of isolated points, continuous curves or connected regions. Selecting image features depend on the problem of image registration and the type of application on it. A feature is defined as an interesting part of an image. Since features are used as the starting point and main primitives for subsequent algorithms. The overall algorithm will often only be as good as its feature detector. Consequently the desirable property for a feature detector is repeatability; whether or not the same feature will be detected in two or more different images of the same scene. Feature detection is a low-level image processing operation. It is usually

80

A New Feature-based Image Registration Algorithm

performed as the first operation on an image and examines every pixel to see if there is a feature present at that pixel. After the strong feature points detected from reference and sensed images, a corresponding mechanism is required between two feature point sets. This correspondence mechanism fulfils the requirement of pairing the feature point of reference image with its correspondent one in the sensed image. References [2-3] are worked on feature point matching using on Zernike moment. It is a very important step because the percentage of correct matches depends on how well the transformation can be estimated in the next step. Common matching methods include simple city block distance, Euclidean distance matching and nearest neighbor based matching. The mapping function is constructed after the feature correspondence has been established. It should transform the sensed image to overlay it over the reference one. The transformation could be rigid, affine, projective or nonlinear [1]. The first broad category of transformation models includes linear transformation which includes translation, rotation, scaling, and other affine transforms. The second category of transformations allows ‘non-rigid’ transformations. These transformations are capable of locally warping the target image to align with the reference image. Non-rigid registration using FFD (Free Form Deformation) shows on [4-5]. The final step of any registration algorithm involves the actual mapping of the one image to other using the transform model estimated in past step. Point-by-point will be corresponding one image to another in one coordinate system. Each pixel from the sensed image can be directly transformed using the estimated mapping functions. Reference [6] published a survey article covering main interpolation methods for re-sampling with the emphasis on medical imaging applications. There are no re-sampling functions necessarily interpolate the image gray levels [7]. This paper is organized as follows: Section 2 discusses

previous literatures. Section 3 presents our developed feature-based image registration algorithm. Section 4 shows experimental results. Section 5 gives conclusions with future works.

2. Literature Review Image registration is widely used in different important areas. References [8-9] were reviewed on medical image registration techniques. A phase-based and some feature-based image registration algorithms were implemented by [10-13]. A new registration algorithm based on Newton-Raphson iteration is proposed to align images with rigid body transformation [14]. PET/CT and MR Image registration using NCC (Normalized Cross Correlation) algorithm and Spatial Transformation (ST) techniques introduced on [5, 15-16]. 3D real-time elastic image registration and multimodal medical image registration algorithm implemented by [6, 17]. Comparisons of edge-based and ridge-based registration of CT and MR brain images done by [18]. Algorithms based on Q-shift, entropy, automatic robust implemented in [19-21]. Reference [22] proposed Invariant Image Recognition by Zernike moment. Image registration using correlation techniques introduced by [23-25]. High resolution remote sensing image registration by combination of feature-based and area-based proposed by [26]. Image and surface registration using neural network implemented by [27-32]. References [33-34] proposed genetic image registration techniques. Non-rigid image matching and registration were implemented from some of author’s [35-38].

3. Proposed Algorithm We are implemented a feature-based image registration algorithm. We used edge as features of image. For detecting edge of image, we used sobel edge detector. Fig. 1 shows the proposed technique for feature-based image registration. Feature extraction involves simplifying the amount of resources required to describe a large set of data

A New Feature-based Image Registration Algorithm

81

pixels with the same label share certain visual characteristics. For intensity images, there are four popular approaches such as threshold techniques, edge-based methods, region-based techniques, and connectivity-preserving relaxation methods. We are applied here edge-based methods for image segmentation that are shown in Fig. 2. We are used to detect edge feature of images using sobel edge detector. The sobel operator performs a 2D spatial gradient measurement on an image. Typically it is used to find the approximate absolute gradient magnitude at each point in an input grayscale image. The sobel edge detector uses a pair of 3×3 convolution kernel, one estimating the gradient in the x-direction and the other estimating the gradient in the y-direction. A convolution kernel is usually much smaller than the actual image. As a result, the kernel is slid over the image and manipulating a square of pixels at a time. The sobel operator consists of a pair 3×3 convolution Fig. 1 Proposed feature-based image registration technique.

accurately. There are many ways to extract features from an image. Low-level features to be those basic features that can be extracted automatically from an image without any shape information. Low level feature is called edge of image and it aims to produce a line drawing. High-level feature extraction concerns finding shapes in an image. Shape extraction implies finding their position, orientation and their size. Image segmentation for the purpose of obtaining boundaries is an important step in a feature-based registration system. There are many segmentation techniques available that could be used. However, there is no unique segmentation technique that can perform best on all types of images, and most segmentation techniques are image-dependent. The goal of segmentation is to simplify and/or change the representation of an image into something that is more meaningful and easier to analyze. It is typically used to locate objects and boundaries in images. More precisely, image segmentation is the process of assigning a label to every pixel in an image such that

kernel is shown in Fig. 3. One kernel is simply rotated by 90° to get another kernel. We used hausdorff distance to measure image difference according to their edges. Hausdorff distance from set A to set B is a maximum minimum function defined as H (A, B) = max a € A {min b € B {d (a, b)}} (1) Where a and b are points of set A and B respectively and d (a, b) is the distance between these points. For simplicity, we took d (a, b) as the Euclidian distance between a and b. Fig. 4 is shown hausdorff distance classifier between two images. Distance = max (H (E1, E2), H (E2, E1)) (2) H (E1, E2) = maxPi€E1 (minQj€E2 (||Pi-Qj||)) (3) Now we are ready to find the matching feature point pairs between f1(x, y) and f2(x, y). The first step of the process is to rotate all the points in f1(x, y) to their new positions located in F1(X, Y) where, X = x cos (θ ) + y sin (θ ) (4)

Y = − x sin (θ ) + y cos (θ ) (5) If an edge point in f1 is rotated to a new position in F1,

the angle θ is also adjusted to suit the coordinates of F1.

A New Feature-based Image Registration Algorithm

82

Fig. 2 Image segmentation.

Fig. 5 Registered images using our algorithm for two images that was taken from different times.

Fig. 3 Sobel convolution kernel.

Fig. 6 Registered images using our algorithm that was taken from different places. Fig. 4 Hausdorff distance of two images.

Reference [39] proposed an iterative scheme that could not remove false pairs completely and efficiently. A new method is proposed to eliminate the incorrectly matched pairs. We present a non iterative scheme based on the idea that the distance between two points in the same image will be preserved when it undergoes a rigid transform. If all the matching pairs are correct then the following equation should hold Pi=Sqi +T for i=1, 2 ….Nm (6) Where S and T are scalar and translation vector respectively.

Fig. 7 Reference image and Scene image (aerial) that taken different times.

4. Experimental Results

Fig. 8 Reference image and scene image (aerial) that taken from different places.

We will see some resulting images that are used in our experiments. Figs. 5-6 are resulting image using our proposed IR technique whereas Figs. 7-8 were the

edge as a feature of an image. There are a lot of image registration systems using other features rather than edge as a feature of image.

reference and scene image respectively.

References 5. Conclusions We used Intel core i3 processor with 4GB RAM to run our experiment. The timing complexity was 1.2 seconds to come output results using MATLAB 2008a. Our proposed technique for image registration is used

[1]

[2]

M.V. Wyawahare, P.M. Patil, H.K. Abhyankar, Image registration techniques: An overview, International Journal of Signal Processing, Image Processing and Pattern Recognition 2 (3) (2009). J. Sarvaiya, S. Patnaik, H. Goklani, Image registration using NSCT and invariant moment, International Journal

A New Feature-based Image Registration Algorithm

[3]

[4]

[5] [6]

[7]

[8]

[9]

[10]

[11]

[12]

[13] [14]

[15]

[16]

[17]

of Image Processing 4 (2) (2010) 119-130. A. Khotanzad, Y.H. Hong, Invariant image recognition by Zernike moment, IEEE Trans. PAMI 12 (5) (1990) 489-497. D. Rueckert, L.I. Sonoda, C. Hayes, D.L.G. Hill, M.O. Leach, D.J. Hawkes, Nonrigid registration using free-form deformations: Application to breast MR images, IEEE Transactions on Medical Imaging 18 (8) (1999) 712-721. C.R.C. Pareja, Real-time 3D elastic image registration, Ph.D. Thesis, Ohio State University, 2004. T.M. Lehmann, C. Gonner, K. Spitzer, Survey: interpolation methods in medical image processing, IEEE Transactions on Medical Imaging 18 (1999) 1049-1075. P.T. Venaz, T. Blu, M. Unser, Interpolation revisited, IEEE Transactions on Medical Imaging 19 (2000) 739-758. J.B.A. Maintz, M.A. Viergever, A survey of medical image registration, Medical Image Analysis 2 (1) (1998) 1-36. R. Shams, P. Sadeghi, R.A. Kennedy, R.I. Hartley, A survey of medical image registration on multicore and the GPU, IEEE Signal Processing Magazine 27 (2) (2010) 50-60. A. Nikaido, K. Ito, T. Aoki, E. Kosuge, R. Kawamata, A phase-based image registration algorithm for dental radiograph identification, in: 2007 IEEE International Conference on Image Processing (ICIP), 2007. K. Krish, S. Heinrich, W.E. Snyder, H. Cakir, S. Khorram, A new feature based image registration algorithm, ASPRS 2008 Annual Conference, Portland, Oregon April 28-May 2, 2008. M.S. Yasein, P. Agathoklis, A feature-based image registration technique for images of different scale, in: IEEE International Symposium on Circuits and Systems (ISCAS 2008), 2008. S. Boda, Feature-based image registration, M. Tech Thesis, National Institute of Technology, 2009. W. Chu, L. Ma, J. Song, T. Vorburger, An iterative image registration algorithm by optimizing similarity measurement, Journal of Research of the National Institute of Standards and Technology 115 ( 1) (2010) 1-6. P.J. Kostelec, S. Periaswamy, Image registration for MRI, Modern Signal Processing, MSRI Publications, vol. 46, 2003. B. Balasubramanian, K. Porkumaran, PET/CT and MR image registration using normalized cross correlation algorithm and spatial transformation techniques, International Journal of Computer and Network Security 2 (5) (2010). A. Almhdie, C. Léger, M. Deriche, R. Lédée, Multimodal medical image registration using a novel implementation of the ICP algorithm, in: 15th European Signal Processing Conference (EUSIPCO 2007), Poznan, Poland, September

83

3-7, 2007. [18] J.B.A. Maintz, P.A.V. Elsenyz, M.A. Viergever, Comparison of edge-based and ridge-based registration of CT and MR brain images, Medical Image Analysis 1 (2) (1996) 151-161. [19] H.S. Own, An efficient image registration algorithm based on q-shift complex wavelet transform (q-shift CWT), ICGST-GVIP Journal 5 (2) (2005) 9-12. [20] M.R. Sabuncu, Entropy-based image registration, Ph.D. Thesis, Department of Electrical Engineering, Princeton University, 2006. [21] G. Yang, C.V. Stewart, M. Sofka, C. Tsai, Automatic robust image registration system: Initialization, estimation, and decision, in: Proceedings of the fourth IEEE International Conference on Computer Vision Systems (ICVS 2006), IEEE, 2006. [22] C. Harris, M. Stephens, A combined corner and edge detector, in: Fourth Alvey Vision Conference, 1988, pp. 147-151. [23] K. Takita, T. Aoki, Y. Sasaki, T. Higuchi, K. Kobayashi, High-accuracy subpixel image registration based on phase-only correlation, IEICE Trans. Fundamentals E86-A (8) (2003) 1925-1934. [24] B. Pan, K. Qian, H. Xie, A. Asundi, Two-dimensional digital image correlation for in-plane displacement and strain measurement: a review, Measurement Science and Technology 20 (6) (2009). [25] W.K. Pratt, Correlation techniques of image registration, IEEE Transactions on Aerospace and Electronic Systems 10 (1974) 353-358. [26] G. Hong, Y. Hang, Combination of feature-based and area-based image registration technique for high resolution remote sensing image, in: IEEE international symposium on Geoscience and Remote Sensing (IGARSS 2007), July 2007, pp. 377-380. [27] L. Ramirez, N.G. Durdle, V.J. Raso, Medical Image Registration in computational Intelligence framework: A review, in: Proceedings of CCECE 2003, May 2003, pp. 1021-1024. [28] T. Sabisch, A. Ferguson, Bolouri, Automatic registration of complex images using a self-organizing neural system, in: Proc. 1998 Int. Joint Conf. on Neural Networks, 1998, pp. 165-170. [29] H. Liua, J.Q. Yan, D. Zhang, Three-dimensional surface registration: A neural network strategy, Neurocomputing 70 (2006) 597-602. [30] M. Li, W. Cai , Z. Tan, A region-based multi-sensor image fusion scheme using pulse-coupled neural network, Pattern Recognition Letters 27 (16) (2006) 1948-1956. [31] L. Shang, J.C. Lv, Z. Yi, Rigid medical image registration using PCA neural network, Neurocomputing 69 (2006) 1717-1722.

84

A New Feature-based Image Registration Algorithm

[32] J. Zhang , Y. Ge , S.H. Ong , C.K. Chui b, S.H. Teoh , C.H. Yan, Rapid surface registration of 3D volumes using a neural network approach, Image and Vision Computing 26 (2007) 201-210. [33] J.J. Jack, C. Roux, Registration of 3-D images by genetic optimization, Pattern Recognition Letters 16 (1995) 823-841. [34] Z. Janko, D. Chetverikov, A. Ekart, Using a genetic algorithm to register an uncelebrated image pair to a 3D surface model, Engineering Applications of Artificial Intelligence 19 (2006) 269-276. [35] R.S. Mitra, N.N. Murthy, Elastic maximal matching, Pattern Recognition 24 (1991) 747-753.

[36] R. Szeliski, S. Lavall, Matching 3-D anatomical surfaces with non-rigid deformations using octree splines, in: IEEE Workshop on Biomedical Image Analysis, 1994, pp. 144-153. [37] W. Peckar, C. Schnorr, K. Rohr, H.S. Stiehl, Two step parameter free elastic image registration with prescribed point displace-ments, Journal of Mathematical Imaging and Vision 10 (1999) 143-162. [38] G. Wollny, F. Kruggel, Computational cost of nonrigid registration algorithms based on fluid dynamics, IEEE Transactions on Medical Imaging 21 (2002) 946-952. [39] B. Zitova, J. Flusser, Image registration methods: a survey, Image and Vision Computing 21 (2003) 977-1000.

Suggest Documents