J. Biomedical Science and Engineering, 2010, 3, 538-542
doi:10.4236/jbise.2010.35075 Published Online May 2010 (http://www.SciRP.org/journal/jbise/
Published Online May 2010 in SciRes. http://www.scirp.org/journal/jbise
Medical equipments high precise detection technology basing
on morphology-harris operator
Yang-Yang Mei, Hai-Ming Xie, Lu Han, Shi-Jun Guo
School of Medical Instrument and Food Engineering, USST,Shanghai, China.
Email: myy19860103@126.com
Received 5 January 2010; revised 8 February 2010; accepted 4 March 2010.
Medical equipments related to life safety of human, it is
important to detect by a high precise method. Image
mosaic which based on Harris corner operator is a
commonly used method in this area; Harris operator
has low calculation burden, it is simple and stable, so it
is more effective comparing with other feature point
extracted operators. But in this algorithm, corner points
can only be detected in a single-scale, there may be los-
ing information of corner points, causing corner point
location offset, extracting false corner points because of
noise. In order to solve this question, the acquired im-
ages should be processed by dilation and erosion opera-
tion firstly, then do image mosaic. Results show that
image noise can be eliminated effectively after those
morphological processes, as well as the false positive
noise generated by image glitch. The success rate of
image mosaic and detection accuracy can be greatly
improved through the Morphology-Harris operator.
Measurement of precision instruments which based on
this new method will improve the measurement accu-
racy, and the research in this area will promote the fur-
ther development of machine vision technology.
Keywords: Image Mosaic; Harris Operator; Feature
Point Extraction; High Precise Detection; Dilation and
With the improvement of living standards, people are
concerned about the health problems more and more,
especially in recent years, surgical accidents have been
happening continuously, which make the physician-
patient relationship nervous. In this situation, the con-
troversy focused on medical equipment whether it is
qualified or not is increasing, so it is important to detect
medical equipment in a high precise and steady way.
Traditional detection methods are inefficiency, the accu-
racy is often affected by Man-made factors. Revolution-
ary changes of precision equipment detection have hap-
pened after the rapid development of computer. Through
computer vision, we can make non-contact detection for
medical instruments which not only improves the detec-
tion accuracy but also reduces the impact of Man-made
factors. However, a number of common machine vision
detection equipments which are restricted by the optical
field of view can not capture large view field and wide
angle images, and if we pursuit large field of view sim-
ply, the image resolution will be reduced, the detection
accuracy will be reduced too. In this case, some re-
quirements of detection accuracy of medical devices
have not been fulfilled. To resolve this problem, this
paper presents a method of using image mosaic which
based on Morphology-Harris operator to obtain images
of the medical instruments, enhancing image resolution
and measuring accuracy.
Image mosaic is a technology which is used to solve the
problem of small view field that cannot make a big pic-
ture. It uses computer to match several images auto-
matically to merge a large panoramic image, at the same
time because of the mosaic of one image from two im-
ages, the image resolution has been greatly improved, so
it has a wide using range in modern life. In foreign
country, the panoramic image mosaic technology which
based on the movement was brought by Richard Szeliskj
in 1996, it is the representative and becomes a classic
image mosaic algorithm [1], Richard Szeliskj became
the founders of the field of image mosaic, he proposed
theory has become a classic theoretical system, even
today there are still many people study his theory of
splicing. On this basis, Shmuel Peleg, Benny Rousso and
others brought an automatic adaptation image mosaic
model in 2000, it will select adaptive model of splicing
in accordance with the different camera movements [2].
Then M. Brown published an article entitled Recongis-
ing Panoramas on the General Assembly of ICCV in
2003 [3], and image mosaic using of SIFT algorithm [4]
Y. Y. Mei et al. / J. Biomedical Science and Engineering 3 (2010) 538-542 539
Copyright © 2010 SciRes. JBiSE
was first brought in this article, the algorithm will be
automatic completed fully and the results are effective. At
the same time, the development of image mosaic is grow-
ing rapidly in our country too, algorithms which were pro-
posed by Xiao-Rui Wang, Zu-Xun Zhang et al. greatly
promoted the development of image mosaic technology.
Image mosaic technology was early used in photo-
graphic mapping to integrate a large number of satellite
images. In this application, the accuracy of splicing does
not ask for much. But, in detection of precision instru-
ments, especially for precision medical equipments, it is
the key point; detection accuracy is affected directly by
it. The image mosaic model is shown in Figure 1. Image
registration is the key of image mosaic; it is closely re-
lated to the success of it. At present, the method of im-
age registration, based on feature point, is used com-
monly and widely. It has several advantages such as low
calculation burden, high sensitive and little affected by
noise [5].
In field of image mosaic, points that change obviously
are called feature points. Extraction of feature points is
the basis of image registration. There is one sort of fea-
ture points called corner points, which are changing in-
tensely of gray value or the intersection point of the out-
line in image, they reflect the important information of
image, so we can get the information that is important
and ignore information that is secondary by corner
points. Harris corner detector is the most commonly
used operator in image mosaic.
3.1. Principle of Harris Corner Operator
Harris detector was brought forward by C. Harris and
M. J. Stephens in 1988 and based on the feature points
of image [6]. While this feature detector is usually
called corner detector, it is not just the corners select-
ing, but rather any image location having large gradi-
ents in all directions at a predetermined scale [7]. Sig-
nals are processed through auto-correlation function
and the result is a matrix M of auto-correlation func-
tion. Eigenvalue of matrix M is first-order curvature of
the correlation function, and the point could be viewed
as a corner if two curvature values are high. Only the
first-order difference of the gray image is used in Har-
ris, so the operation is simple and is widely used in
practice. The principle is to calculate the gray value
changes during the window moving along any direc-
tion; it is supposed that the small window of center
pixel point has moved along the direction )( yx,u
Figure 1. Image mosaic model.
, as well as along the direction of , then
analytical expression about gray value changes will be
given out by Harris, as Eq.1:
 
,, ,,
EuvwxyI xuyvI xy
In the function above, stands for the win-
dow function;
 
xu Ixyyv
stands for
the gradient of gray image; usually, stands
for gauss filter. For a small displacement, Eq.1 can be
replaced by Eq.2, as follows:
MvuvuE ],[),( (2)
can be expressed as follows:
xy y
In Eq.3,
and are the first derivatives of each
image channel. is very similar to the local auto-
correlation function and
describes the shape of the
origin of auto-correlation function. 1
, 2
are as-
sumed to be two eigenvalues of
, and they will
proportion with the local auto-correlation function,
constituting a non-variable rotation to
[8]. Through
judgment 1
and 2
the region which changed
slowly could be determined, as well as corner and
The maximum value of local region can be defined
by Harris feature point:
TMdenotes the matrix trace of
, and T(M) =
DM stands for the determinant of matrix
. The value of is usually recom-
mended of 0.02 - 0.06.
With the theory of Harris corner detection, several
steps could be concluded as follows:
1) Filter every pixel of the image by horizontal and
vertical difference operator, through this,
can be obtained [9].
2) Deal with the four elements of matrix
gauss filter, a new
will be obtained. Gaussian
function is showed below:
exp 2
Image Registeration Image fusion
Image input
3) Calculate the interesting value of each pixel by
540 Y. Y. Mei et al. / J. Biomedical Science and Engineering 3 (2010) 538-542
Copyright © 2010 SciRes. JBiSE
function Eq.4.
4) Set the threshold and do non-maximum suppres-
sion by it.
The corner points, detected by Harris detection are
shown in Figure 2.
3.2. Disadvantage of Harris Operator and
Ameliorate Method
Harris operator is a widely used feature point extraction
operator and has several advantages, but it has some
disadvantages too. For example, in this algorithm, cor-
ner points can only be detected in a single-scale, there
may be losing information of corner points, causing
corner point location offset, extracting false corner
points because of noise, inaccurate localization to T
and diagonal T and so on.
For tiny medical devices, measurement accuracy is
extremely demanding, but because of light, as well as
the lens distortion, the image edges will produce some
distortion. If the device edge which near the image
edge has a glitch, the glitch will have a shape distortion
in the two images for mosaic, causing corner point lo-
cation offset, thus affecting the accuracy of image mo-
saic. For which we first using morphological dilation
and erosion operation to preprocess the image, elimi-
nating the false positive noise generated by image
glitch, thus eliminating the inaccurate positioning of
the corner points caused by noise, as well as the effect
of the corner point location offset, greatly improved the
splicing accuracy and detection accuracy.
4.1. Dilation and Erosion Operation
According to the illumination of 3.2, first, we process
the image through dilation and erosion operation. The
result is shown in Figure 3, the left part is the former
image, we can see the glitch clearly in the small circle
marked on the image, and right half part is image after
processing, the glitch is completely disposed.
4.2. Image Mosaic
In this part, we use tablet punch to demonstrate the mo-
saic course. The size of tablet punch determines the
weight and density of tablets, the precise of punch is
important. First of all, two images of tablet punch using
a large field camera can be acquired, then do dilation and
erosion operation, the result can be shown in Figure 4.
Be attention that the two images must have some over-
lapped parts, and this is the basis of image mosaic.
Then we can extract the feature points of each image
by Harris operator; do image registration by the nor-
malized cross-correlation and image fusion. The result is
shown in Figure 5.
Figure 2. Corner points detected by Harris.
Figure 3. Dilation and erosion processing.
Figure 4. Images waiting for mosaic.
Figure 5. Image after mosaic.
4.3. Test and Contrast
After image dilation and erosion processing, mosaic
images through registration and fusion, then the width of
tablet punch is measured. Here, we test the width in
three situations: single image, mosaic image without
preprocess, mosaic image by Morphology-Harris Opera-
tion. The result images are shown in Figure 6, the left is
single image, middle is non-process image result, the
right one is mosaic image through Morphology-Harris
operator. We found the accuracy of the width in the right
Y. Y. Mei et al. / J. Biomedical Science and Engineering 3 (2010) 538-542 541
Copyright © 2010 SciRes.
one is the highest, the left one is the lowest. Size is
measured by a method which based on sub-pixel meas-
urement, the pixel value is not an integer number in this
way and the measurement result has a higher precise; the
value at top left corner of image is every pixel value
which is measured after the camera is calibrated using
the 30 mm standard block .
In order to compare the result conveniently, we can
see the Table 1 below. In this table, five values can be
measured through moving the high of camera in every
situation. Absolute difference between single image size
and standard size is 0.567 mm; the value between non-
preprocess mosaic image and standard size is 0.430 mm;
and the value between Mosaic image of morphological
processes and standard size is 0.080 mm. Comparing the
results, it is obvious to see that the measurement accu-
racy of mosaic image after morphological processes is
much higher than the single image and Mosaic image of
non-preprocess. Through this comparison, the important
of dilation and erosion processing of image mosaic in
the precision instrument measurement is proved, and
the detection accuracy will be improved through im-
age mosaic.
Medical equipments related to life safety of human, ac-
curate measurement is particularly important. In this
paper, we use the morphology-Harris operator preproc-
essing images, eliminate the noise caused by lens distor-
tion and light uneven, extract feature point accurately,
position accuracy is higher than the untreated image.
Through dilation and erosion processes, the accuracy of
the mosaic image is improved, simultaneity the cost of
additional hardware needed is reduced too. The experi-
ment results has proved that, accuracy of medical device
testing which based on morphology-Harris operator of
image mosaic fully meets the requirements. This is an
efficient, high-precision medical device detection tech-
nology, suitable for promotion.
Figure 6. Width of tablet punch image.
Table 1. Result compared.
pixel value actual value of every
pixel actual value of width average value of width
740.178 0.02072 mm 15.336 mm
765.842 0.02003 mm 15.340 mm
690.223 0.02222 mm 15.337 mm
801.132 0.01899 mm 15.210 mm
Single image
787.045 0.01949 mm 15.341 mm
15.313 mm
979.082 0.01578 mm 15.450 mm
901.110 0.01715 mm 15.453 mm
1008.037 0.01532 mm 15.437 mm
880.029 0.01756 mm 15.450 mm
Mosaic image of
947.501 0.01632 mm 15.461 mm
15.450 mm
1000.250 0.01578 mm 15.784 mm
921.283 0.01715 mm 15.800 mm
1031.527 0.01532 mm 15.803 mm
899.544 0.01756 mm 15.796 mm
Mosaic image of mor-
phological processes
969.301 0.01632 mm 15.819 mm
15.800 mm
Standard size ----- ----- 15.880 mm 15.880 mm
542 Y. Y. Mei et al. / J. Biomedical Science and Engineering 3 (2010) 538-542
Copyright © 2010 SciRes. JBiSE
Here, I wish to thank my tutor Professor Haiming Xie, under his care-
ful guidance, I completed the paper and related experiments, thanks my
lab colleagues Lu Han, Shijun Guo, and under their help I finished my
dissertation fast and accurately.
[1] Szeliski, R. (1996) Video mosaics for virtual environments.
IEEE Computer Graphics and Applications, 16(2), 22-30.
[2] Peleg, S. and Rousso, B. (2000) Mosaicing on adaptive
manifolds. IEEE Transactions on PAMI, 22(10), 1144-1154.
[3] Brown, M. and Lowe, D.G. (2003) Recognising pano-
ramas. Proceedings of IEEE International Conference on
Computer Vision, 2(1), 1218-1225.
[4] Mikolajczyk, K. and Schmid, C. (2004) Scale & Affine
invariant interest point detectors. International Journal of
Computer Vision, 60(1), 63-86.
[5] Xie, D.H. and Zhan, Z.Q. (2003) Improving Harris cor-
ner detector. Journal of Geomatics, 28(2), 22-23.
[6] Harris, C. and Stephens, M.J. (1988) A combined corner
and edge detector. Proceedings Fourth Alvey Vision Con-
ference, Manchester, 1988, 147-151.
[7] Qian, W., Fu, Z.Z. and Liu, Q.Q. (2008) Voting-
strategy-based approach to image registration. Opto-Elec-
tronic Engineering, 35(10), 86-91.
[8] Zhao, W.J., Gong, S.R. and Liu, C.P. (2008) Adaptive
Harris corner detection algorithm. Computer Engineering,
34(10), 212-214.
[9] Zhang, J.F. and Xing, X. (2008) Static image registration
research based on Harris corner. Science and technology
information, 20, 9-10.