Phương pháp đếm giấy dựa vào giá trị chênh lệch mức xám

perform edge 
enhancement on gratscale images
The calculation of the grayscale 
value are made according to the line 
gray curves
Set the binarization threshold, 
Change the difference gray into 
rectangular image
Determine the number of sheets 
based on the number of rectangular 
image output the result
8NGHIÊN CỨU KHOA HỌC
Tạp chí Nghiên cứu khoa học,Trường Đại học Sao Đỏ, ISSN 1859-4190 Số 4 (67).2019
pattern recognition, and computer vision. Since 
the 1990s, researchers have conducted in-depth 
research on the application of Gabor filters in 
texture segmentation, object detection, image 
enhancement, and image feature extraction. The 
commonly used even-symmetric two-dimensional 
Gabor filter can be expressed as:
 (2)
 (3)
Where: 
θ is the direction of the Gabor filter;
σu and σv are the standard deviations of the Gaussian envelope on the u-axis and the v-axis 
(u-axis is parallel to θ, v-axis is straight to θ), and 
ω is the modulation frequency.
First, a Gabor filter template with different spatial 
frequencies of 5 × 5 is designed. Since the Gabor 
function is a Gaussian function modulated by a 
complex sine function, when filtering the paper, 
the real part of the filter is used to filter the paper. 
Define the image for filtered by the Gabor filter as
 (4)
In the formula: i (x, y) is the input grayscale image, 
* represents the convolution operation.
3.3. Gray projection
Gray projection includes row projection and column 
projection. Row projection is obtained by adding 
the gray values of each point in each row in the 
image. Column projection is obtained by adding 
the gray values of each point in each column in 
the image. The projection value of each row or 
column is described in a rectangular coordinate 
system, and a gray-scale projection curve is 
obtained. The gray-scale projection curve can be 
used to intuitively see the distribution of gray-scale 
information in the image. There is a noticeable 
difference in the stripe gray level of paper, so it 
can be used as a method of counting paper. In this 
study, in order to reduce the amount of calculation 
and keep the pixel value between 0 ~ 255, the 
average value of the gray value of each column of 
pixels was used for calculation.
In the formula: row (i) represents the sum of pixels 
in i-th row of the image, and col (j) represents the 
sum of pixels in the j-th column of the image.
3.4. Gray difference calculation
Perform gray-level difference calculation on 
the gray-scale projected image, that is, the sum 
of the gray values of the projected image in the 
projection area is calculated as a one-dimensional 
vectơ, and the starting value of the statistical 
gray-value vectơ is set to 0, and the subtraction 
operation is performed on the grayscale values in 
the order of backward difference in order to obtain 
another one-dimensional vectơ, and the value of 
this vectơ is the result obtained by the difference 
of the grayscale values. In this paper, the results 
obtained after graying the original image and 
projecting the difference are shown in Fig. 3.
It can be seen from Fig.3 that the above-mentioned 
method for performing differential operation on 
gray values has a wave shape trajectory locally, 
which corresponds to the light and dark stripes 
of the paper in Fig. 1. It can be seen that when 
the corresponding gray value at the edge of the 
paper is small and the corresponding gray value is 
large at the gap between the two papers, the wave 
trough in the curve obtained after the differential 
calculation of the image corresponds to the two 
papers and the peaks correspond to the paper, 
so the peaks and valleys will appear periodically 
in the curve after the differential calculation of the 
image. The gray difference between the adjacent 
peaks and valleys corresponds to the white stripes 
to black in the image. Stripe jump, this jump can be 
considered as the dividing line between two sheets 
of paper. Then, based on this characteristic, an 
algorithm can be designed to calculate the number 
of sheets.
3.5. Paper counting algorithm
The paper counting algorithm works as follow:
1) In the effect map obtained by differentiating 
the gray projection curve, find all the peaks and 
troughs in the curve, then find the length of the 
interval between two adjacent peaks or troughs, 
the last all the peak-to-peak intervals or valley 
intervals are counted.
2) Perform a binarization operation according to 
the difference projection map, the specific method 
is: if the value of the gray value at a certain point 
changes from less than 0 to greater than 0, it is 
assumed that there is 1 sheet of paper here, the 
transition points greater than 0 are marked with 
red rectangles in the original image.
3) From the first point of the peak or trough, 
determine whether it is within the range of [len-wid, 
len + wid], where wid is a constant that is customized 
to reduce the counting error. In this study, wid = 3. 
If it is within this range, the next judgment is made; 
otherwise, the point is discarded.
h(x, y) = !"#$!$" exp *− !" ,%#$!# + &#$"#./ cos(ωu) u = xcosθ + ysinθ, v = −xcosθ + y sin θ° 
f° = |i(x, y)∗h(x, y)|° 
 (5)
 (6)
row(i) = (img(i, j)!"#$ 
 col(j) = (img(i, j)°&'#$ 	 
9LIÊN NGÀNH ĐIỆN - ĐIỆN TỬ - TỰ ĐỘNG HÓA
Tạp chí Nghiên cứu khoa học, Trường Đại học Sao Đỏ, ISSN 1859-4190 Số 4 (67).2019
Fig.4. The paper counting algorithm flow chart
Table 1. Comparison of this method and method [10]
a
b
Fig.5. The counting renderings from two original 
graphs
4) Determine whether the pixel width between the 
binarized waveforms is greater than 1.5 * len. If 
yes, it indicates that a paper is added between the 
two, otherwise, the next judgment is made.
5) The paper is counted according to the number 
of marked red rectangles, and the paper counting 
effect is displayed with a red rectangle in the 
grayscale image.
The specific process of the paper counting 
algorithm is shown in Fig. 4.
4. EXPERIMENT AND RESULT ANALYSIS 
In order to verify the accuracy of the counting paper 
using the grayscale method, this paper counts the 
paper under two different visual effects in Fig.1, and 
displays the counting effect. Because the paper 
counting method used in this paper depends to a 
large extent on the nature of the difference curve, 
and is calculated from the peak or valley of the 
curve after determining the properties of the apex 
and valley. According to the above-mentioned 
paper counting algorithm, the difference curves of 
the paper are counted according to two different 
visual effects in Fig. 1. The result count is shown 
in Fig.5. It is calculated according to Fig. 5. that 
the accuracy of the counting paper calculated 
according to the difference gray value exceeds 
98%.
From Table 1, it can be concluded that for two 
different paper imaging effects, the accuracy of 
the gray value difference algorithm proposed in 
this paper is relatively high. At the same time, 
Method
Image (a) Image (b)
Actual 
value
Test 
value
False 
detection/
missed 
detection
Accuracy Actual value
Test 
value
False 
detection/
missed 
detection
Accuracy
Method [10] 65 61 4 93.84% 70 65 5 92.85%
Proposed 
method 65 64 1 98.46% 70 70 0 100%
Select a certain height in the original 
image and sum the image gray values Initialization, paper quantity is 0
Draw the summed gray value curve
Differential operation on adjacent gray 
pixels
Preliminary determination of the pixel 
width of a single paper based on the peak 
and valley characteristics of the gray curve
Determine whether the pixel width 
between waveforms after binarization 
is between [len-wid, len+wid]
Yes
Yes
Number of sheet + 1
Determine whether the pixel width 
between waveforms after binarization 
is greater than 1.5 len
No
No
The number of sheet +1, mark their 
positions
Count the total number of papers and 
output the results
10
NGHIÊN CỨU KHOA HỌC
Tạp chí Nghiên cứu khoa học,Trường Đại học Sao Đỏ, ISSN 1859-4190 Số 4 (67).2019
the accuracy of the method used in this paper 
has reached more than 98%, especially in the 
original image b. Although there are two papers 
stuck together, proposes algorithm is still able to 
accurately identify and detect the number of paper, 
which is greatly improved and improved compared 
to the method of edge extraction directly using 
Gabor filters. The reason why the algorithm 
used in this paper is better than the algorithm 
that directly applies the Gabor filter to the paper 
edge extraction is because in this algorithm, 
the properties between adjacent stripes of the 
paper are analyzed and processed, and the gap 
between papers The accurate positioning was 
performed, the number of false detections and 
missed detections was reduced, and the counting 
accuracy was improved.
5. CONCLUSION
Based on the analysis of the above experimental 
results, it is known that in the case of neatly 
arranged papers, based on obtaining a light and 
dark stripe image, this paper uses a method based 
on gray value difference to count the number of 
papers. Under the condition that the gray value 
is differentiated, the fundamental reason why 
this paper counting method works well is that the 
difference method can better extract the stripe 
characteristics of the paper, because the light and 
dark stripes are the fundamental basis for judging 
the number of papers, that is, by the gray level 
difference of the black and white stripes realizes 
the analysis of the image characteristics, and at 
the same time, the number of papers is obtained 
by accurately positioning the peaks or valleys in 
the difference curve.
The paper counting algorithm based on gray 
difference can realize the function of paper 
counting, but there are still some problems in this 
new paper counting method, such as the counting 
accuracy in the case of large differences in paper 
thickness and uneven alignment between papers, 
etc. This will be the focus of the next research.
REFERENCES
[1] Jian Gao, Xiao Wang An Apparatus and Method 
for Stacked Sheet Counting with camera Array, 
IEEE Latin merica Transactions 2013, 978-1-
4799-0333-7
[2] J. G. A. Barbedo, A Review on Methods for 
Automatic Counting of Objects in Digital Images, 
IEEE Latin America Transactions 2012, 10, (5), 
2112-2124.
[3] Junya Sato, Toru Koide, Yuki Kisi, Basic Study 
on Facial Oil Blotting Paper Counting Using a 
Webcam.IEEE Transactions on Instrumentation 
and Measurement 18 October 2018.
[4] Mato, J. L.; Alvarez Souto, M.; Besteiro, R.; 
Moledo,J.A, Automated counting of palletized 
slate slabs based on machine vision, in Proc. 39th 
Annu. Conf. IEEE Ind. Electron. Soc. (IECON), 
Vienna, Austria 2013, pp. 2378-2383.
[5] Auboussier,E.; Berthe, B.; Fumey,T., Device for 
counting stacked products. U.S. Patent May 16, 
2006, (7045 765).
[6] Perdoux, D.; Bachar Bittar; Gilles Leroux S.A., 
Device for counting products stacked side-by-
side, U.S. Patent Nov. 11, 1997, 5 686 729.
[7] Chatchai Suppitaksakul; Rattakorn, M., Machine 
vision system for counting the number of 
corrugated cardboard. In Proc. Int. Elect. Eng. 
Congr, (IEECON), pp. 1-4. 2014.
[8] Chen T.; Wang. Y.N; Xiao C.Y., An Apparatus and 
Method for Real-Time Stacked Sheets Counting 
With Line-Scan Cameras, IEEE Transactions 
on Instrumentation and Measurement 2015, 
(64(7):1876-1884), 1876-1884.
[9] Hong Zhao; Rong D; Xiao, ChangYan, A Machine 
Vision System for Stacked Substrates Counting 
With a Robust Stripe Detection Algorithm, IEEE 
TRANSACTIONS ON SYSTEMS, MAN, AND 
CYBERNETICS: SYSTEMS 2017, 2168-2216 c 
2017 IEEE., (10.1109/TSMC.2017.2766441).
[10] Li Yi, Ruan Qiuqi. Algorithm of Paper Counting 
Based on Texture Analysis[J]. Journal of Image 
and Graphics, 2004, 9(9), 1042-1048.
Pham Thi Dieu Thuy
- Pham Thi Dieu Thuy received the B.S. degree in Automation from Thai Nguyen 
University of Technology, Thai Nguyen, Viet Nam
+ In 2006 and the M.S. degree in Measurement and Control systems from Ha Noi 
University of Science and Technology, Ha Noi, Viet Nam
+ In 2010: Her current research interests include medical image processing, machine 
vision and machine learning
- Email: dieuthuy303@gmail.com
- Telephone No: 0986468005
AUTHORS BIOGRAPHY
11
LIÊN NGÀNH ĐIỆN - ĐIỆN TỬ - TỰ ĐỘNG HÓA
Tạp chí Nghiên cứu khoa học, Trường Đại học Sao Đỏ, ISSN 1859-4190 Số 4 (67).2019
Ha Minh Tuan
- Ha Minh Tuan received the B.S. degree in Automation from Viet Nam Maritime 
University, Hai Phong, Viet Nam
+ In 2005 and the M.S. degree in Measurement and Control systems from Ha Noi 
University of Science and Technology, Ha Noi, Viet Nam
+ In 2010: His current research interests include medical image processing, machine 
vision and machine learning
- Email: minhtuanha031@gmail.com
- Telephone No: 0977536826
Luong Thi Thanh Xuan
- Luong Thi Thanh Xuan received the B.S. degree in Technology Education from Thai 
Nguyen University of Technology, Thai Nguyen
+ In 2003 and the M.S. degree in Automation from Thai Nguyen University of Technology, 
Thai Nguyen, Viet Nam
 + In 2011: Her current research interests include Control Engineering and Automation. 
Summary of current work: Lecturer, Faculty of Electrical Engineering, Sao Do University
- Email: thanhxuan7980@gmail.com
- Telephone No: 0982791980
Changyan Xiao 
- College of Electrical and Information Engineering, Hunan University, Changsha, China
- Changyan Xiao received the B.Eng. and M.S. degrees in mechanical and electronic 
engineering from the National University of Defense Technology, Changsha, China, 
in 1994 and 1997, respectively, and the Ph.D. degree in biomedical engineering from 
Shanghai Jiao Tong University, Shanghai, China, in 2005.
- He was a Visiting Post-Doctoral Researcher with the Division of Image Processing, 
Leiden University Medical Center, Leiden, The Netherlands, from 2008 to 2009. Since 
2005, he has been an Associate Professor and a Full Professor with the College of 
Electrical and Information Engineering, Hunan University, Changsha. His current 
research interests include medical imaging, machine vision, and embedded instrument.
- Email: c.xiao.@hnu.edu.cn
Nguyen Thi Viet Huong
- Nguyen Thi Viet Huong received the B.S. degree in Electrical Engineering Technology 
from Hung Yen University of Technology and Education, Viet Nam, in 2009 and the M.S. 
degree in Control Engineering and Automation from Hung Yen University of Technology 
and Education,Viet Nam, in 2014. Her current research interests include Control 
Engineering and Automation. Summary of current work: Lecturer, Faculty of Electrical 
Engineering, Sao Do University.
- Email: nguyenthiviethuong1986 @gmail.com
- Telephone No: 0911311086