Locating MRI discrepancy by orientation invariant method
Introduction
In the past twenty years, researchers invested a lot of effort to develop different techniques to identify the difference of images. This past work includes – principal component analysis [4], [25], [26], moment invariant [20], trace transform [19], fusion code of real part and imaginary part [15], Eigenspace [8], Limbus and black purple detection [6], [18], Contour angle histogram [1], R-Theta transform [2], Delaunay tessellation [3], calculation of rotating angle [7], [25], 3D to 2D remainder [9], positive correlation [10], 3D to 2D approach [11], [12], congruent transform [5], [14], [23], area discrimination [13], [14], [16], contour following, U-turn feature [17], hook up the missing stroke [21], baseline estimation [22], sharp U-turn extraction [24].
Here, we propose a different approach, which combines several techniques, to cope with MRI image shifting and rotating problems. Thus, after pictures of the MRI images are taken, the algorithms developed here can automatically locate the discrepancy of MRI images without involving further human effort. Fig. 1 shows different kinds of MRI images. At beginning, MRI image digitizer-machine digitizes MRI image and then MRI image is transferred to the BMP file. In this research, there are two kinds of MRI images. The first kind of image is called healthy MRI image. The second kind of image is called illness MRI image. The healthy and illness images are shown in Fig. 6. The discrepancy of these two kinds of images can be detected by the technique developed in this research. Since the backgrounds are irrelevant to find the orientation and centroid of the image, backgrounds are removed from the MRI images. The result is shown in Fig. 2. Major axis and centroid algorithms are used to find orientations and centroids of these two images. Both algorithms can very precisely locate the centroids and orientations of these MRI images. This research first shifts healthy MRI image to make the centroid of healthy MRI image to overlap with centroid of illness MRI image. Then, healthy MRI image is rotated. Thus, healthy MRI image can be aligned to same orientation as illness MRI image. After this, image subtraction can be applied to these two MRI images. The subtracted result can very precisely show the discrepancy of these two MRI images. Under normal circumstance, when MRI images are taken, even though MRI image are a little shifted or a little rotated, the algorithms developed in this research will still correctly find the discrepancy of healthy and illness MRI images.
This paper consists of five sections. Section 2 discusses how to extract the important feature of MRI images. Section 3 explains how to find centroids and orientations of MRI image. Section 4 performs image subtraction. Section 5 concludes this paper.
Section snippets
Discard the background of MRI image
In order to find locations and orientations of the MRI images, the important feature of MRI images must be extracted. Edge tracking algorithm is used to locate edges of MRI images and backgrounds are removed from MRI images. MRI images with no background are shown in Fig. 2.
Finding the centroid
The sizes of the images used in this research are M ∗ N image array. The centroid (xc, yc) of the right image in Fig. 2 can be found by Eqs. (1), (2). In Eqs. (1), (2), the term g(x, y) is the gray level of the pixel at the location (x, y).
Using second moment to find orientation of MRI image
The major axis is the axis around which the object will have the minimum moment of inertia. This is useful in determining the object’s orientation. Fig. 3 shows the object major axes. Fig. 4 shows relative position of major axis. The major axis and X-axis generate
MRI image shifting and rotating
Using the method described in the previous section, the orientation of each extracted MRI image can be obtained. In this research, there are two kinds of images: the perfect (healthy) and the flawed (illness) images. The healthy and illness images are shown in Fig. 6. Darker images are illness images. As mentioned before, in order to overlap healthy image with illness image, Healthy image needs to be shifted and rotated. Fig. 5 shows corresponding position after healthy image is rotated θ
Results and conclusions
Since one can find both centroid and orientation of MRI image, by using previous mentioned rotating, transferring, and interpolating technique, illness MRI image can be aligned to the same orientation as healthy image. The rotated and shifted images are shown in Fig. 8. After transferring and subtraction is performed to both images. The subtracted image is show in Fig. 9. By this result, one can find that the technique developed in this research can work well to find the discrepancy between
References (26)
Automatic identification for Chinese seal image
Pattern Recognition
(1996)- et al.
Automatic Chinese seal identification
Computer Vision Graphics and Image Processing
(1984) - et al.
Biometric iris recognition system using a fast and robust iris localization and alignment procedure
Optics and Lasers in Engineering
(2006) - et al.
Three-dimensional approach to thresholding seal impressions
Pattern Recognition
(1996) - et al.
An automatic seal imprint verification approach
Pattern Recognition
(1995) - et al.
Palmprint identification using feature-level fusion
Pattern Recognition
(2006) - et al.
Unconstrained seal imprint verification using attributed stroke graph matching
Pattern Recognition
(1989) - et al.
Hand biometrics
Image and Vision Computing
(2006) Registration of seal images using contour analysis
- et al.
Seal identification using the Delaunary tessellation
Proceedings of the National Science Council, ROC (A)
(1998)