Network Pharmacology, 2020, 5(1-2): 1-32

Article

imageProcAnal: A novel Matlab software package for image

processing and analysis

WenJun Zhang
School of Life Sciences, Sun Yat-sen University, Guangzhou 510275, China; International Academy of Ecology and
Environmental Sciences, Hong Kong
E-mail: zhwj@mail.sysu.edu.cn, wjzhang@iaees.org

Received 8 November 2019; Accepted 23 December 2019; Published 1 June 2020

Abstract
In present study, I developed a powerful Matlab-based software package, imageProcAnal (Version 1.0), for
image processing and analysis. Several modules were available for uses. Functions to resize, crop, rotate, dilate,
pixelate and watermark images are included in Basic module. Adjustment of image edge, contrast and gamma
factor is available in Adjustment module. In Denoising module, four elementary denoising methods and their
joint uses are provided for image filtering and denoising. In Edge and Contour Detection module, five methods
for detection of image edge and contour are provided. Also three methods for image sharpening are available
in Sharpening module. Segmentation module includes three methods for image segmentation. Connected
components recognition can be made in Object Recognition module. In Analysis module, Pearson correlation
measure was provided to align images. Pearson correlations and p values among images can be obtained.
Mutiple methods across different modules can be sequentially and jointly used to achieve ideal image. Finally,
some examples for image processing can be found in Demo module. In addition to conventional methods, I
proposed two new methods for image segmentation, i.e., TDOS (Two-Dimensional Ordered Segmentation)
and MWAS (Moving Windows Averaging Segmentation). In TDOS, which is based on two-dimensional
ordered cluster analysis, the pixels belonging to the same segment (class) are adjoint and are not separated by
any of remaining segments. MWAS is based on moving windows averaging method for transect boundary
detection. Algorithms of TDOS and MWAS were given, and full Matlab codes of MWAS were given.

Keywords Matlab; software; image denoising; image adjustment; edge detection; image sharpening; image
segmentation; image alignment; Pearson correlation.

Network Pharmacology
ISSN 2415­1084
URL: http://www.iaees.org/publications/journals/np/online­version.asp
RSS: http://www.iaees.org/publications/journals/np/rss.xml
E­mail: networkpharmacology@iaees.org
Editor­in­Chief: WenJun Zhang
Publisher: International Academy of Ecology and Environmental Sciences

1 Introduction
Image recognition and processing is a discipline that uses computer systems to identify and process images to
achieve a human-like visual system that understands the outside world (Wu, 2006; Bhardwaj and Solanki,

IAEES www.iaees.org
2 Network Pharmacology, 2020, 5(1-2): 1-32

2016). An image is an information carrier that conveys information to people in different patterns. Some of the
patterns can be recognized and understood by computer vision systems by simulating the visual function of the
human eye, but due to insufficient prior knowledge in related fields, some other patterns cannot be simulated
and identified by computer vision systems. The factors that affect the development of image recognition and
processing research are mainly the prior knowledge of related fields. This includes not only existing computer
algorithms, but also knowledge of image acquisition, image-related databases, etc.
Image processing and analysis is a fundamental aspect for various sciences and technologies, for example,
medical digonosis, gene expression, landscape recognition, etc (Dogra et al., 2018; Zhang and Qi, 2019).
However, the common used software for image processing are somewhat simple and short of many
functionalities. The present study aims to develop and present the powerful Matlab-based software for image
processing and analysis. Various functionalities seldom found in common software are available in the
software pakage. In addition, new methods are proposed also.

2 Materials and Methods

2.1 Methods
Various methods for image processing and analysis, and corresponding Matlab codes are presented in the
following.
2.1.1 Rotate, crop, resize, dilate and watermark images
Rotate an image with a specified degree:

degr=10;
rgb=imread('geneExpression.png');
j=imrotate(rgb,degr, 'bilinear', 'crop'); %Rotate image by degr degrees
imshow(j);

Crop the image with a specified rectangular area:

rgb=imread('geneExpression.png');
rgb=imcrop(rgb); %Crop image interactively
imshow(rgb);

Resize the image to ba pixels:

a=600; b=800;
rgb=imread('geneExpression.png');
j=imresize(rgb, [a b]) %Resize image to ba pixels
imshow(j);

Dilate the image:

m=5;
rgb=imread('wjzhang.png');
imj=im2bw(rgb);
se=strel('diamond',m);
j=imdilate(imj,se) %Dilate image with a diamond of size m

IAEES www.iaees.org
 Network Pharmacology, 2020, 5(1-2): 1-32 3

imshow(j);

The general process of embedding watermark in an image using discrete cosine transform is: (1) Discrete
cosine transformation of the original image; (2) generation of watermark signals; (3) embed watermark in the
image. The following is the procedure for the inverse of two-dimensional discrete cosine transformation to
obtain a watermarked image:

rgb=imread('wjzhang.png');
imdim=numel(size(rgb));
if (imdim~=2) rgb=rgb2gray(rgb); end
alfa=0.01; len=1000; %Set parameters
dctf=dct2(rgb);
[m,n]=size(dctf); %DCT (Discrete Cosine Transformation) of the image
watermark=randn(len,1);
[y0,z0]=sort(watermark); %Generate watermark series and sort it
v=dctf(:);
[y1,z1]=sort(v);
mn=m*n;
k=len;
p=zeros(mn,1); %Find the position for embeding watermark
for i=1:mn
if (k>=1)
p(mn)=y1(mn)*(1+alfa*y0(k));
k=k-1;
else p(mn)=y1(mn);
end
mn=mn-1;
end
y=zeros(mn,1);
mn=m*n;
for i=1:mn
y(z1(i))=p(i);
end
s=1;
for j=1:n
for i=1:m
dctf2(i,j)=y(s);
s=s+1;
end
end %Revise the amplitudes of s frequency components with greater amplitudes and embed watermark
idctf=idct2(dctf2);
imshow(idctf,[]); %Inverse of DCT for obtaining the image embedded with watermark

2.1.2 Image adjustment

IAEES www.iaees.org
4 Network Pharmacology, 2020, 5(1-2): 1-32

Image adjustment is a processing method that highlights certain information in an image according to specific
requirements, such as improving some edge information and adjusting the contrast of the image, while
weakening or removing some unwanted information. The adjusted image can often enhance the ability to
recognize special information, improve the visual effect, and allow the observer to check a more direct and
clear image.
2.1.2.1 Grayscale linear transformation
Insufficient brightness or non-linearity of some images will make the contrast of images less than ideal. We
can improve the contrast of an image by reassigning the pixel amplitude. The grayscale of the image was
expanded by linear mapping. In addition, the image compression method is adopted, or the grayscale is
segmented, that is, compression is performed in a certain interval and expanded in another interval according
to image characteristics and our requirements.
2.1.2.2 Gamma adjustment
Gamma value in the function ‘imadjust’ can determine the correction effect. Gamma value determines the
grayscale mapping method of the input image to the output image, that is, determines whether to adjust low
grayscale or adjust high grayscale. When Gamma equals 1, it is a linear transformation.
2.1.2.3 Two-dimensional filtering
The spatial texture information of an image can reflect the position, shape, size and other characteristics of the
image. Using linear filtering technology can adjust some texture features of the image and remove other
unuseful features. Linear filtering is a neighborhood operation whose response is given by the sum of the
product of the filter coefficients and the corresponding pixel values of the area swept by the filtering window.
Matlab provides a variety of filtering operators, such as mean filtering, Gaussian lowpass filtering, Laplacian
operator, motion blur operator, edge adjustment, edge extraction, contrast enhancement filter, etc.
The following is the full Matlab codes for image adjustment:

rgb=imread('geneExpression.png');
imdim=numel(size(rgb));
if (imdim~=2) i=rgb2gray(rgb);
else i=rgb;
end
h=fspecial('prewitt');
j=imfilter(i,h); %Edge adjustment
imshow(j);

a=0.6; b=0.9; c=0.2; d=0.5;

rgb=imread('geneExpression.png');
imdim=numel(size(rgb));
if (imdim~=2) i=rgb2gray(rgb);
else i=rgb;
end
j=imadjust(i,[a b],[c d]); %Contrast adjustment; Grayscale from [a b] to [c d]
imshow(j);

gamma=2;
rgb=imread('geneExpression.png');
imdim=numel(size(rgb));

IAEES www.iaees.org
 Network Pharmacology, 2020, 5(1-2): 1-32 5

if (imdim~=2) i=rgb2gray(rgb);
else i=rgb;
end
j=imadjust(i,[],[], gamma); %Gamma adjustment
imshow(j);

2.1.3 Image denoising

There may be a variety of noises in an image. These noises may be generated during propagation, or they may
be generated during image processing such as quantization. In Matlab there are roughly five types of noises:
'gaussian' is Gaussian noise, 'localvar' is Gaussian white noise, 'poisson' is Poisson noise, 'salt & pepper' is salt
and pepper noise, and 'speckle' is uniform random noise.
Restoration of noisy image is an important area of image processing. The negative effect of these noises on
the image can be eliminated by smooth filtering. A good smoothing method should eliminate these noises
without blurring the edges and lines of the image. There are many filtering methods, such as mean filtering,
median filtering, adaptive filtering, wavelet denoising, etc. Each method has its own advantages and
disadvantages.
2.1.3.1 Mean filtering
Mean filtering belongs to the spatial filtering and smoothing technology. It mainly uses the neighborhood
averaging method to remove the particle noise obtained by scanning the image. The neighborhood averaging
method mainly uses the average value of several neighboring pixels to replace a certain pixel.
2.1.3.2 Median filtering
Median filtering is a commonly used non-linear smoothing filtering. Its filtering principle is similar to the
mean filtering. The difference between the two is that the output pixel value of the median filtering is
determined by the intermediate value of the neighboring pixels instead of the average value. Median filtering is
very effective for filtering the salt and pepper noise of the image, and the noise on the image is almost
completely removed.
2.1.3.3 Adaptive filtering
The Matlab image processing toolbox provides ‘wiener2’ to estimate the local mean and variance of each pixel
to achieve adaptive filtering of image noise.
2.1.3.4 Wavelet denoising
There are three main steps in image denoising using wavelet analysis: (1) wavelet decomposition of the image
signal; (2) threshold quantization of the decomposed high-frequency coefficients, and (3) image signals are
reconstructed using two-dimensional wavelets.
The following is the full Matlab codes of jointly used four methods above for image denoising:

rgb=imread('geneExpression.png');
imdim=numel(size(rgb));
if (imdim~=2) i=rgb2gray(rgb);
else i=rgb;
end
h=fspecial('average');
j=imfilter(i,h); %Mean filtering
k=medfilt2(j); %Median filtering
rgb=wiener2(k); %Adaptive filtering
rgb=im2double(rgb);

IAEES www.iaees.org
6 Network Pharmacology, 2020, 5(1-2): 1-32

thr=0.1; sorh='s'; crit='shannon'; keepapp=0;

rgb=wpdencmp(rgb,sorh,3,'sym4',crit,thr,keepapp); %Wavelet denoising
imshow(rgb);

2.1.4 Edge detection

2.1.4.1 Edge detection
Edge information is an important attribute for extracting image features. The edge detection operators include
Prewitt operator, Canny operator, Sobel operator, Log operator, etc.
The edge detection of the image using several edge detection operators in MATLAB is as follows:

rgb=imread('geneExpression.png');
imdim=numel(size(rgb));
if (imdim~=2) i=rgb2gray(rgb);
else i=rgb;
end
bw1=edge(i,'sobel'); %Sobel method
bw2=edge(i,'roberts'); %Roberts method
bw3=edge(i,'prewitt'); %Prewitt method
bw4=edge(i,'log'); %Laplacian of Gaussian method
bw5=edge(i,'canny'); %Canny method
subplot(2,3,1);imshow(bw1);title('Sobel edge detection');
subplot(2,3,2);imshow(bw2);title('Roberts edge detection ');
subplot(2,3,3);imshow(bw3);title('Prewitt edge detection ');
subplot(2,3,4);imshow(bw4);title('Laplacian of Gaussian edge detection ');
subplot(2,3,5);imshow(bw5);title('Canny edge detection ');

2.1.4.2 Contour detection

The following is the Matlab codes for contour detection of a binary image:

rgb=imread('wjzhang.png');
rgb=im2bw(rgb);
stru=strel('diamond',1);
imd=imdilate(rgb,stru);
ime=imerode(rgb,stru);
imed=imd-ime;
contour=1-imed;
imshow(contour);

2.1.5 Image sharpening

Image sharpening is mainly used to ajdust the edges and changing parts of grayscale of an image. There are
two processing methods, spatial domain and transformation domain. The image sharpening operators include
Roberts operator, Sobel operator, Prewitt operator, Laplacian operator, etc.

rgb=imread('geneExpression.png');

IAEES www.iaees.org
 Network Pharmacology, 2020, 5(1-2): 1-32 7

imdim=numel(size(rgb));
if (imdim~=2) i=rgb2gray(rgb);
else i=rgb;
end
subplot(2,2,1);imshow(i);title('Original');
h1=fspecial('sobel');
i1=filter2(h1,i);
subplot(2,2,2);imshow(i1);title('Sobel Sharpening ');
h2=fspecial('prewitt');
i2=filter2(h2,i);
subplot(2,2,3);imshow(i2);title('Prewitt Sharpening');
h3=fspecial('laplacian');
i3=filter2(h3,i);
subplot(2,2,4);imshow(i3);title('Laplacian Sharpening');

2.1.6 Image segmentation

Image segmentation refers to the separation of different areas with special significance in an image. These
areas are disjoint from each other. Each area meets certain similarity criteria such as grayscale, texture, and
color.
2.1.6.1 Maximum Inter-segment Variance Threshold
The maximum inter-segment variance threshold method proposed by Ostu is derived on the basis of least
square method. The following is to segment an image using maximum inter-segment variance threshold
method.

thr=128;
rgb=imread('geneExpression.png');
imdim=numel(size(rgb));
if (imdim~=2) c=rgb2gray(rgb);
else c=rgb;
end
subplot(1,2,1);imshow(c);
title('Original Image');
freq=imhist(c);
[r,t]=size(c);
N=r*t;
L=256;
freq=freq/N;
for i=2:L
if freq(i)~=0
st=i-1;
break
end
end
for i=L:-1:1
if freq(i)~=0;

IAEES www.iaees.org
8 Network Pharmacology, 2020, 5(1-2): 1-32

nd=i-1;
break
end
end
f=freq(st+1:nd+1);
p=st;q=nd-st;u=0;
for i=1:q;
u=u+f(i)*(p+i-1);
ua(i)=u;
end
for i=1:q
w(i)=sum(f(1:i));
end
d=(u*w-ua).^2./(w.*(1-w));
[y,tp]=max(d);
th=tp+p;
if th<=210
th=tp+p;
else th=210;
end
rgb1=zeros(r,t);
for i=1:r
for j=1:t
X1(i,j)=double(c(i,j));
end
end
for i=1:r
for j=1:t
if (X1(i,j)>=thr)
rgb1(i,j)=X1(i,j);
else
rgb1(i,j)=0;
end
end
end
subplot(1,2,2);imshow(rgb1);title('Segmentated Image');

2.1.6.2 Single Threshold Histogram

If the histogram of the image shows obvious double peaks, the grayscale corresponding to the valley between
the double peaks is selected as the threshold for segmentation, that is, the single threshold histogram
segmentation.

thr=128;
rgb1=imread('geneExpression.png');
imdim=numel(size(rgb1));

IAEES www.iaees.org
 Network Pharmacology, 2020, 5(1-2): 1-32 9

if (imdim~=2) rgb=rgb2gray(rgb1)
else rgb=rgb1;
end
[m,n]=size(rgb);
for i=1:m
for j=1:n
if (rgb(i,j)>thr);
rgb(i,j)=255;
else
rgb(i,j)=0;
end
end
end
subplot(1,2,1);imshow(rgb1); title('Original Image');
subplot(1,2,2);imshow(rgb); title('Segmentated Image');

2.1.6.3 Two-dimensional ordered segmentation (TDOS)

Two-dimensional ordered cluster analysis (Zhang, 1993; Zhang et al., 2014) can be used for image
segmentation. In this method, the pixels belonging to the same segment (class) are adjoint and are not
separated by any of remaining segments. As a new method for image segmentation, here I name it as TDOS
(Two-Dimensional Ordered Segmentation) method.
Assume there are p indices and q pixels (q=mn, i.e., the total number of elements (pixels) of mn
grayscale matrix). Now we have xi=(xi1, xi2,…, xip), i=1,2,...,q. For gray image segmentation, p=1, i.e.,
grayscale value; for RGB color image segmentation, p indices can be represented by grayscale, red value,
green value, blue value, etc.
In the ordered cluster analysis, the pixels in the same segment are spatially adjacent, and the lines between
any two pixels that belong to different segments are not intersected.
The following is the method from Zhang (1993). Assume the two-dimensional coordinate, i.e., the row and
column IDs, of pixel i is (yi1, yi2), i=1,2,…,q.
(1) For pixels xi and xj, i=1,2,...,q-1; j>i, calculate

G1(i,j)={xk, xi | yk2>( yj2- yi2)( yk1- yi1)/( yj1- yi1)+ yi2}

G2(i,j)={xk, xj | yk2<( yj2- yi2)( yk1- yi1)/( yj1- yi1)+ yi2}
‫ܩ‬ҧ 1(i,j)={xk, xj | yk2>( yj2- yi2)( yk1- yi1)/( yj1- yi1)+ yi2}
‫ܩ‬ҧ 2(i,j)={xk, xi | yk2<( yj2- yi2)( yk1- yi1)/( yj1- yi1)+ yi2}

Thus there will be q(q-1) possible segmentaions. Calculate the error function for each segmentaion:

e(G(2))=∑2S=1D(iS, iS+1 -1)

where D(i, j)= ∑2S=1(xS-‫ݔ‬ҧ i-j)(xS-‫ݔ‬ҧ i-j)’, and

‫ݔ‬ҧ i-j=(∑jS=i xS )/(j-i+1), i<j, i=1,2,...,q-1

IAEES www.iaees.org
10 Network Pharmacology, 2020, 5(1-2): 1-32

Choose the segmentaions that meets min e(G(2)), the two segments can be achieved.
(2) Assume that k segments have been achieved. Following the procedure above, segmentate them as two
segments respectively, and choose the segmentation that meets max min e(G(2)), thus k+1 segments is
achieved.
(3) If k<q, segmentation continues, and if k=q, the calculation terminates.
Error function can be used as cluster distance level. At given level, we can use all resulatant segments as
the segments of the image.
The improved two-dimensional ordered clustering method (Zhang et al., 2014) is also available for TDOS
method.
2.1.6.4 Moving windows averaging segmentation (MWAS)
Moving windows averaging segmentation is based on boundary detection algorithm (Zhang and Schoenly,
1999; Zhang, 2005). Here I name it as MWAS.
For a gray image, the grayscale value, j, of any pixel falls in [0,255]. Calculate the frequency distribution
of grayscale values. It means that there are 256 ordered grayscale values along the x-axis, and 1 measurement
variable at each grayscale value j (i.e., number of the pixels with grayscale value j, fj).
Consider a window with window width Q in calculation of moving average segmentation. This window is
split into two equal halves, Wa and Wb, each half has Q/2 ordered grayscales. The grayscale location of this
window is defined by k+0.5, where k=Q/2,Q/2+1,Q/2+2,...,256-Q/2. The average of fj in each half window is

Wa(k+0.5)= ∑kj=k-Q/2+1fj/(Q/2)
Wb(k+0.5)= ∑k+Q/2j=k+1fj/(Q/2)

The distance between each of the resulting 256-Q+1 pairs of averages can be calculated. For a window,
calculate the absolute distance

DS(k+0.5) =|Wa(k+0.5)-Wb(k+0.5)|

Monte Carlo technique is used to estimate expected mean distance of distance matrix for a window width
Q. Each grayscale value is randomly repositioned along grayscale series 0~255 for l=1,2, .. ,p times (e.g.,
p=200), and distance for each of the reorderd data sets, producing a new distance matrix DR(k+0.5,l). For each
k+0.5 window midpoint grayscale along the series (0,1,2,…,255), the mean expected distance and standard
deviation are as follows

DSB(k+0.5)= ∑pl=1(DR(k+0.5,l)/p
SD(k+0.5)=(∑pl=1(DR(k+0.5,l)-DSB(k+0.5))2/(p-1))0.5

The overal expected mean distance and average standard deviation for the series at window width Q is
given by

DSB(·)= ∑256- Q/2k=Q/2 DSB(k+0.5)/(256-Q+1)

SDB= ∑256- Q/2k=Q/2 SD(k+0.5)/(256-Q+1)

For each window midpoint location, the distance estimate can be transformed into Z score as follows

IAEES www.iaees.org
 Network Pharmacology, 2020, 5(1-2): 1-32 11

DZ(k+0.5)=(DS(k+0.5)-DSB(·))/SDB

As done in Zhang et al. (2005), If the present ratio SDB2/DSB(·) against the cumulative ratios of all
window widths is less than 0.05, i.e., 95% of information has been obtained, or the number of window widths
is greater than n-Q+1, then start to calculate pooled distance.turn to pooling procedure.
Finally, the pooled distance estimate is calculated for each window midpoint grayscale from T different
window widths

DZB(k+0.5)= ∑Ts=1 DZ(k+0.5,s)/T

The overal expected mean distance and average standard deviation for the series of pooled results are PDSB=0
and PSDB=1 respectively.
Calculate lower limit of the confidential interval with different confidence degree (Zhang and Schoenly,
1999)

PDSB -2* PSDB (Grain Level I)

PDSB -1.5* PSDB (Grain Level II)
PDSB -1.0* PSDB (Grain Level III)
PDSB -0.5* PSDB (Grain Level IV)

If the pooled distance, DZB(k+0.5), of window midpoint is less than the upper limit, we consider the
corresponding grayscale location is a vally grayscale.
After doing these, the boundary grayscale locations (valley grayscale locations) can be determined across
the series 0~255. Thus, the interval [0,255] can be segmentated into several segments or no any segment.

mw=3; st=9; %Initial window width=3; Increment of window width=9

percent=5; %Cumulative information=95%
sim=200; %Monte Carlo simulations=200
rgb=imread('brainTumorMRI.png');
imdim=numel(size(rgb));
if (imdim~=2) rgb=rgb2gray(rgb);
else rgb=rgb;
end
freq=imhist(rgb);
s=1; %1 variable, i.e., grayscale
a(s,:)=freq';
ts=256;
kp=1;
prop=0;su=0;
sm=0; w=mw;
while (sm>=0)
sm=sm+1;
t1=0; t2=0;
z=0;
for i=1:ts-w+1

IAEES www.iaees.org
12 Network Pharmacology, 2020, 5(1-2): 1-32

for k=1:s
for j=i:w-1+i
if (j>(floor(w/2)+z)) t2=t2+a(k,j); else t1=t1+a(k,j); end
end
b(k,i)=t1/floor(w/2);
b(k,i+1)=t2/floor(w/2);
t1=0; t2=0;
end
md(i,sm)=floor(w/2)+z+0.5;
z=z+1;
ds2(i,sm)=0;
for l=1:s
ds2(i,sm)=ds2(i,sm)+abs(b(l,i)-b(l,i+1));
end
end
for i=1:ts-w+1
dsb(i)=0;
p(i)=0;
qq(i)=0;
end
kk=1;
while (kk<=sim)
cols=randperm(ts);
for k=1:s
for j=1:ts
cc(k,j)=a(k,cols(j));
end
end
t1=0; t2=0; z=0;
for i=1:ts-w+1
for k=1:s
for j=i:w-1+i
if (j>(floor(w/2)+z)) t2=t2+cc(k,j); else t1=t1+cc(k,j); end
end
b(k,i)=t1/floor(w/2);
b(k,i+1)=t2/floor(w/2);
t1=0; t2=0;
end
z=z+1;
ds=0;
for k=1:s
ds=ds+abs(b(k,i)-b(k,i+1));
end
dsb(i)=dsb(i)+ds/sim;
p(i)=p(i)+ds^2;

IAEES www.iaees.org
 Network Pharmacology, 2020, 5(1-2): 1-32 13

qq(i)=qq(i)+ds;
end
kk=kk+1;
end
for i=1:ts-w+1
sd(i)=(p(i)-2*qq(i)*dsb(i)+sim*dsb(i)^2)/(sim-1);
if (sd(i)>=0) sd(i)=sqrt(sd(i)); else sd(i)=0;
end
end
momn=0; mosd=0;
for i=1:ts-w+1
momn=momn+dsb(i);
mosd=mosd+sd(i);
g(i)=ds2(i,sm);
end
mnbar(kp)=momn/(ts-w+1);
sdbar(kp)=mosd/(ts-w+1);
su=su+sdbar(kp)^2/mnbar(kp);
prop=sdbar(kp)^2/mnbar(kp)/su;
for i=1:ts-w+1
dz(i,sm)=(ds2(i,sm)-mnbar(kp))/sdbar(kp);
end
kp=kp+1;
if (prop<=(percent/100.0))
q=kp-1; break;
end
if (sm>=ts-w+1)
q=kp-1; break;
end
w=w+st;
end
wc=floor(st/2)*q-floor(st/2)+ts-(mw+st*(q-1))+1;
hk=ts-(mw+st*(q-1))+1;
s2=floor(st/2)*q-floor(st/2)+1;
mt=1;
tv=0;
for i=1:ts-mw+1
if (tv>=hk) mt=mt-1; end
if ((i>=s2) && (tv<hk))
mt=q; tv=tv+1;
end
pdz(i)=0;
ap=i;
for j=1:mt
pdz(i)=pdz(i)+dz(ap,j);

IAEES www.iaees.org
14 Network Pharmacology, 2020, 5(1-2): 1-32

ap=ap-floor(st/2);
end
pdz(i)=pdz(i)/mt;
s1=i*1.0/floor(st/2);
s3=(i-wc)*1.0/floor(st/2);
if ((floor(s1)==s1) && (i<s2)) mt=mt+1; end
if ((s3>floor(s3)) && (i>wc)) mt=mt+1; end
end
for i=1:ts-mw+1
g(i)=pdz(i);
end
mnbarr=0; sdbarr=1;
for j=1:4
th=0;
switch (j)
case 1
th=-2*sdbarr; %Deep vally grayscale detection: I
case 2
th=-1.5*sdbarr; %Mediate vally grayscale detection: II
case 3
th=-1*sdbarr; %Mild vally grayscale detection: III
case 4
th=-0.5*sdbarr; %Shallow vally grayscale detection. IV
end
on=1; off=1; ids=0;
for i=1:ts-w+1
if (g(i)<=(mnbarr+th))
if ((on==1) | (off==0))
ids=ids+1;
segs(j,ids)=i;
on=0; off=1;
end
continue;
else
if ((off==1) | (on==0))
off=0; on=1;
end
end
end
len(j)=ids;
end
segs=segs-1;
s=0;
for k=1:4
if (len(k)==0) continue; end

IAEES www.iaees.org
 Network Pharmacology, 2020, 5(1-2): 1-32 15

s=s+1;
end
[m,n]=size(rgb);
ss=0;
for k=1:4
if (len(k)==0) continue; end
rgb1=zeros(m,n);
ss=ss+1;
if (len(k)==1)
if (segs(k,1)==0)
for i=1:m
for j=1:n
rgb1(i,j)=0;
end
end
end
if (segs(k,1)>0)
for i=1:m
for j=1:n
if ((rgb(i,j)<=segs(k,1)))
rgb1(i,j)=0; end
if ((rgb(i,j)>segs(k,1)))
rgb1(i,j)=255; end
end
end
end
end
if (len(k)>=2)
if (segs(k,1)==0)
for i=1:m
for j=1:n
if ((rgb(i,j)>=segs(k,1)) && (rgb(i,j)<segs(k,2))) rgb1(i,j)=0; end
for l=2:len(k)-1
if ((rgb(i,j)>segs(k,l)) && (rgb(i,j)<=segs(k,l+1)))
rgb1(i,j)=segs(k,l); end
end
end
if ((rgb(i,j)>segs(k,len(k))) && (rgb(i,j)<=255)) rgb1(i,j)=255; end
end
end
if (segs(k,1)>0)
for i=1:m
for j=1:n
if ((rgb(i,j)>=0) && (rgb(i,j)<segs(k,1))) rgb1(i,j)=0; end
for l=1:len(k)-1

IAEES www.iaees.org
16 Network Pharmacology, 2020, 5(1-2): 1-32

if ((rgb(i,j)>segs(k,l)) && (rgb(i,j)<=segs(k,l+1)))

rgb1(i,j)=segs(k,l); end
end
end
if ((rgb(i,j)>segs(k,len(k))) && (rgb(i,j)<=255)) rgb1(i,j)=255; end
end
end
end
switch (s)
case 1
subplot(1,1,1);imshow(rgb1);
case 2
subplot(1,2,ss);imshow(rgb1);
case 3
subplot(1,3,ss);imshow(rgb1);
case 4
subplot(2,2,ss);imshow(rgb1);
end
switch (k)
case 1
title('Grain Level I','FontSize',7);
case 2
title('Grain Level II','FontSize',7);
case 3
title('Grain Level III','FontSize',7);
case 4
title('Grain Level IV','FontSize',7);
end
end

2.1.7 Recognition of connected components

Connectedness and connected components are important topological property and structure in topological
spaces (Zhang, 2012, 2016, 2018). Choose a certain pixel of a connected component, dialate it and intersect
with the original image, and iterate until it does not change to obtain the extracted component. By doing so, all
connected components can be obtained. The following Matlab codes are used to recognize connected
components in a binary image:

rgb=imread('brainTumorMRI.png');
rgb=im2bw(rgb);
xr=bwlabel(rgb,8); %Connectedness
u=unique(xr); %max(u): number of connected components
stru=strel('diamond',1); %Diamond structure
for i=1:max(u(:))
ind=find(xr==i); %Choose an object
lind=ind(1);

IAEES www.iaees.org
 Network Pharmacology, 2020, 5(1-2): 1-32 17

x0=zeros(size(rgb));
x0(lind)=1;
x=imdilate(x0,stru);
x1=x0;
while (~isequal(x1,x))
x1=x;
x=imdilate(x1,stru).*logical(rgb);
end
imr(:,:,i)=x;
end
for i=1:size(imr,3)
figure
im=imr(:,:,i);
imshow(im,[]);
end

2.1.8 Image alignment

Suppose we want to align two gray images, represented by two matrices A=(aij) and B=(bij), i=1, 2, ..., m; j=1,
2, ..., n, where m and n are the number of pixels along y-axis and x-axis, 0aij1 and 0bij1 are the
standardized grayscale values at pixel (i, j) of two images respectively. First, two images are transformed to
gray images. Resize the two images to the same dimension of mn, where m and n are the means of
dimensions of two images respectively. The grayscale values of images are standardized as values in [0,1]
(Zhang and Qi, 2019).
Matrices A and B are transformed to vectors A1 and B1 respectively, i.e., (A1, B1)={(ai, bi) | i=1, 2, ..., p}
(Zhang, 2018b). By using Pearson correlation (Zhang, 2012, 2015, 2018a-b; Zhang and Li, 2015), we can
achieve the test results for between-image correlation. A greater Pearson correlation means the higher
similarity in image structure between two images.
Full Matlab codes for Pearson correlation between two images are as follows (Zhang, 2012, 2015, 2018a-b;
Zhang and Qi, 2019):

sig=input('Input significance level(e.g., 0.01): ');

RGB1=imread('geneExpression.png'); RGB2=imread('geneExpression.png');
Gray1=rgb2gray(RGB1); Gray2=rgb2gray(RGB2);
m=round((size(Gray1,1)+size(Gray2,1))/2); n=round((size(Gray1,2)+size(Gray2,2))/2);
I=imresize(Gray1,[m n]); J=imresize(Gray2,[m n]);
A=im2double(I); B=im2double(J);
A1=reshape(A, m*n, 1); %Transform matrix A to vector A1
B1=reshape(B, m*n, 1); %Transform matrix B to vector B1
strr='';
strr=strcat(strr,[char(13) 'Pearson correlation between two images' char(13)]);
r=corr(A1,B1);
strr=strcat(strr,[char(13) 'Pearson correlation r=' num2str(r) char(13)]);
tvalue=abs(r)/sqrt((1-r^2)/(m*n-2));
p=(1-tcdf(tvalue,m*n-2))*2;
sigma=p<sig;

IAEES www.iaees.org
18 Network Pharmacology, 2020, 5(1-2): 1-32

if (sigma==1) strr=strcat(strr,[char(13) 'Pearson correlation is statistically significant (p=' num2str(p) ')' char(13)]); end
if (sigma==0) strr=strcat(strr,[char(13) 'Pearson correlation is not statistically significant (p=' num2str(p) ')' char(13)]); end
fprintf(strr);

2.2 Software package

The software package, imageProcAnal (Version 1.0), and the future versions can be freely downloaded at:
http://www.iaees.org/publications/software/index.asp
Before using the software, users are strongly to install the MATLAB software in your computer. After than,
unrar the software package and double-click imageProcAnal.exe to open the software (Fig. 1). Detailed
methods used during image processing will be recorded above the current image.

Fig. 1 The software, imageProcAnal (Version 1.0). Here an image has been loaded into the window.

2.3 Demo images

Demo images in present article are SF-RT gene expression pattern in S. furcifera, geneExpression.png (Lin et
al., 2019; Fig. 2), MRI image used for experiment is infected with a tumor region in the temporal lobe,
brainTumorMRI.png (Dogra et al., 2018; Fig. 3), and a human image, wjzhang.png (Fig. 4).

IAEES www.iaees.org
 Network Phharmacology, 20
020, 5(1-2): 1-332 19

Fig. 2 SF-RT gene exxpression patterrn in S. furciferra, geneExpresssion.png (Lin ett al., 2019).

Fig. 3 MR
RI image used for
f experiment is infected withh a tumor regio
on in the tempooral lobe, brainTumorMRI.png
g (Dogra et al.,,
2018).

Fig. 4 A human image, wjzhang.png.

IAEES www.iaees.org
w
20 Network Phharmacology, 20
020, 5(1-2): 1-332

3 Softwaare Guide
3.1 File menu
m (Fig. 5)
5
Openn: once the sooftware startss and before making any operation on image, an im mage must bee loaded intoo
the window interface by using Opeen sub-menu..
Reoppen: during anny operation, Reopen can be used to ree-loaded the original
o imagee initially loaaded by Openn
sub-menuu.
Save as: at any steep, the currennt image(s) inn the window can be savedd using Save aas sub-menu.
Save: the filenamee most recenttly used to saave image can n be used agaain to save new image for replacing thee
previous image.
Closee: terminate and
a close the software.

Fig. 5 File meenu.

3.2 Basicc module (Seection 2.1.1; Fig. 6)

Resizze: Resize an image to a sppecified size (Fig. 7a).
Rotatte: Rotate an image by a specified degrree (Fig. 7b).
Crop: Crop a rectaangular area of
o the image (Fig. 8).
Zoomm: Zoom on, off o and out ann image.
Dilatte: Dilate an image
i with sppecified struccture and size (Fig. 9).
Pixellate: Pixelate an image witth the mouse--specified areea (Fig.10). The
T last point is fixed withh double clickk,
and the other
o points are
a fixed with single click.
Wateermark: Wateermark an im mage with speecified param meters (Fig. 111). The copyrright of the image
i can bee
guarantedd by adding the
t watermarkk.

IAEES www.iaees.org
w
 Network Phharmacology, 20
020, 5(1-2): 1-332 21

F 6 Basic mo
Fig. odule.

(a) (b))
Fig. 7 Parameter inpput window for Resize
R (a) and Rotate
R (b).

Fig. 8 Crop a rectangular area

a of the imagge.

IAEES www.iaees.org
w
22 Network Phharmacology, 20
020, 5(1-2): 1-332

Figg. 9 Dilate an im
mage with speciified structure and
a size.

Fiig. 10 Pixelate an image with mouse-specifie

m d area.

Figg. 11 Watermarkk an image with

h specified paraameters.

3.3 Adjuustment mod dule (Section 2.1.2; Fig. 12)

Edgge Adjustmennt: grayscale adjustment
a off image edgess (Fig. 13).
Conntrast Adjustm
ment: grayscaale adjustmennt of an imagee with specifiied contrast pparameters (Fig. 14).
Gammma Adjustm ment: grayscale adjustmeent of an im mage with sppecified gam mma value in n the Matlabb
function imadjust (Figg. 15).

IAEES www.iaees.org
w
 Network Phharmacology, 20
020, 5(1-2): 1-332 23

Fig. 12 Adjustmentt module.

Figg. 13 Original (lleft) and edge adjusted

a (right) images.

Fig. 14 A
Adjustment of an
a image with specified
s contraast parameters.In this example,, the image wass adjusted for more
m darkness.

IAEES www.iaees.org
w
24 Network Phharmacology, 20
020, 5(1-2): 1-332

Fig. 15 Oriiginal (left) andd gamma adjusted (gamma=0.55; right) imagess.

3.4 Denooising modulle (Section 2..1.3; Fig. 16))

It is neceessary to use the methods in this moduule to removee noises in thee images withh backgroundd noises, e.g.,,
the image geneExpresssion.png (Fig. 2). For higgh-quality and d clean imagees as brainTuumorMRI.png g (Fig. 3) andd
wjzhang..png (Fig. 4), the use of thhis module is unnecessary.
Meaan Filtering: see
s section 2..1.3.1.
Meddian Filteringg: see section 2.1.3.2.
Adaaptive Filterinng: see section 2.1.3.3.
Wavvelet Denoisiing: see sectioon 2.1.3.4.
Joinnt (M_M_A) Filtering: Baased on the orriginal image,, sequentiallyy use Mean Fiiltering, Med
dian Filtering,,
and Adaaptive Filterinng to removee noises in thhe image. It is one time and conveniient use of th he first threee
methods..
Joinnt (M_M_A__W) Filteringg: Based on the original image, sequuentially use Mean Filterring, Mediann
Filtering, Adaptive Filtering
F andd Wavelet Deenoising to remove
r noises in the immage. It is one
o time andd
convenieent use of all of
o four methoods.

Figg. 16 Denoising module.

IAEES www.iaees.org
w
 Network Phharmacology, 20
020, 5(1-2): 1-332 25

(a) (b)
Fig. 17 Coomparison betw
ween original annd noise-removved images. (a) Original imagee (Fig. 2); (b) N
Noise-removed image
i by using
Joint (M_M_A) Filtering.

3.5 Edgee and Contou ur Detection module (Secction 2.1.4; Fig.F 18)
Detectionn of object edges
e in an image
i and im mage contou ur can be connducted by uusing the metthods in thiss
module.
Sobbel, Prewitt, Laplacian
L of Gaussian, Rooberts and Caanny Edge Detection:
D onee of these meethods can bee
chosen too detect edges (Section 2.11.4.1; Fig. 199). For an uncclean image with
w noises, liike geneExpreession.png inn
Fig. 2, it is suggested using denoissing methods to remove no oises in advannce, and then make edge detection.
d
Conntour Detectioon: the contouur of the imaage is detectedd using this method
m (Sectiion 2.1.4.2; Fig. 20).

Fig. 18 Edgee and Contour Detection

D moduule.

IAEES www.iaees.org
w
26 Network Pharmacology, 2020, 5(1-2): 1-32

Fig. 19 Canny (upper) and Sobel (lower) edge detection on the images without (left) and with (right) removal of noises (here
using Joint (M_M_A) Filtering).

Fig. 20 Contour detection on the images without (left) and with (right) removal of noises (here using Joint (M_M_A) Filtering).

3.6 Sharpening module (Section 2.1.5; Fig. 21)

One of the methods, Sobel Sharpening, Prewitt Sharpening, and Laplacian Sharpening can be used to sharpen
an image (Fig. 22). For an unclean image with noises, like geneExpression.png in Fig. 2, it is suggested using
denoising methods to remove noises and then make sharpening.

IAEES www.iaees.org
 Network Phharmacology, 20
020, 5(1-2): 1-332 27

Fig.. 21 Sharpening
g module.

wer) sharpeningg of the images without (left) and

Fig. 22 Laaplacian (upper)) and Sobel (low a with (right)) removal of noises (here
using Joinnt (M_M_A) Filltering).

mentation moodule (Sectioon 2.1.6; Fig.. 23)

3.7 Segm
One of the methodss, Maximum m Inter-segmeent Variancee Threshold (Seection 2.1.6.1; Fig. 24),
2 Movingg
Windows Averaging Segmentation (Section 2..1.6.4; Fig. 25-26),
2 and Single Threshhold Hist (Secction 2.1.6.2;;
Fig. 27) can be usedd to make seggmentation on
o an image (Fig. 23). Foor an uncleann image withh noises, likee
geneExpression.png ini Fig. 2, it is suggestedd using deno
oising methoods to removve noises and d then makee
segmentaation.

IAEES www.iaees.org
w
28 Network Phharmacology, 20
020, 5(1-2): 1-332

Fig. 23
2 Segmentatio
on module.

Fig. 24 Maximum Inter-ssegment Variannce Threshold seegmentation meethod (Here thee original imagee, geneExpressiion.png, has
been denoised by using Joint (M_M_A) Filtering).

Fig.. 25 Moving Windows

W Averagging Segmentatiion method (forr brainTumorM
MRI.png).

IAEES www.iaees.org
w
 Network Phharmacology, 20
020, 5(1-2): 1-332 29

Fig. 26 Moving Windows Avveraging Segmeentation methodd (for wjzhang.ppng).

Fig. 27 Sinngle Threshold Hist segmentattion method (H

Here the originall image, geneExxpression.png, hhas been denoissed by using
Joint (M_M
M_A) Filteringg).

3.8 Objeect Recognitiion module (Section

( 2.1.77; Fig. 28)
Once Coonnected Com mponents Reccognition is performed, connected compponents in thhe image will be displayedd
one by onne in differennt windows (F
Fig. 29).

Fig. 28 Object
O Recogniition module.

IAEES www.iaees.org
w
30 Network Phharmacology, 20
020, 5(1-2): 1-332

Fig. 299 One of the connnected componnents in the imaage, brainTumo

orMRI.png (dennoised by using Joint (M_M_A
A) Filtering).

3.9 Anallysis module (Section 2.1.8; Fig. 30)

Currentlyy, Alignmentt method (Zhaang and Qi, 2019)
2 for Peaarson correlattion between images is included in thee
module. By opening several
s imagees, the Pearsoon correlation
n matrix of thhese images ccan be obtain
ned using thiss
method (Zhang
( and Qi,
Q 2019).

Figg. 30 Analysis module.

m

3.10 Dem mo module (FFig. 31)

In this module,
m we cann choose to observe
o variouus demos of image
i processsing (Fig. 322-33).

IAEES www.iaees.org
w
 Network Phharmacology, 20
020, 5(1-2): 1-332 31

F 31 Demo module.
Fig. m

Fig. 32 Demoo for Denoising and Segmentattion.

Fig. 33
3 Demo for Sh
harpening.

IAEES www.iaees.org
32 Network Pharmacology, 2020, 5(1-2): 1-32

References
Bhardwaj N, Solanki A. 2016. An efficient algorithm for color image segmentation. Selforganizology, 3(3):
87-99
Dogra J, Prashar N, Jain S, Sood M. 2018. Improved methods for analyzing MRI brain images. Network
Biology, 8(1): 1-11
Lin J, Zhang M, He JY, et al. 2019. Characterization of two non-LTR retrotransposons from Sogatella
furcifera and Nilaparvata lugens. Arthropods, 8(1): 7-16
Wu HX. 2006. Recognition and Processing of Ecological Landscape Images. MS Dissertation, Sun Yat-sen
University, Guangzhou, China
Zhang WJ. 1993. Two-dimensional order clustering and its application in the regionalization of agriculture.
Bulletin of Soil and Water Conservation, 13(1): 34-41
Zhang WJ. 2005. An improved algorithm on boundary detection of ecological transect and network software.
Journal of Biomathematics, 20(4): 477-486
Zhang WJ. 2012. Computational Ecology: Graphs, Networks and Agent-based Modeling. World Scientific,
Singapore
Zhang WJ. 2015. Calculation and statistic test of partial correlation of general correlation measures.
Selforganizology, 2(4): 65-77
Zhang WJ. 2016. Detecting connectedness of network: A Matlab program and application in tumor pathways
and a phylogenic network. Selforganizology, 3(4): 117-120
Zhang WJ. 2018a. Fundamentals of Network Biology. World Scientific Europe, London, UK
Zhang WJ. 2018b. Network matrix based methods for between-network comparison. Network Biology, 8(4):
144-152
Zhang WJ, Qi YH. 2019. Matlab methods for calculation of between-image correlations and similarities.
Ornamental and Medicinal Plants, 3(3-4): 6-12
Zhang WJ, Li X. 2015. General correlation and partial correlation analysis in finding interactions: with
Spearman rank correlation and proportion correlation as correlation measures. Network Biology, 5(4):
163-168
Zhang WJ, Qi YH. 2019. Matlab methods for calculation of between-image correlations and similarities.
Ornamental and Medicinal Plants, 3(3-4): 6-12
Zhang WJ, Qi YH, Zhang ZG. 2014. Two-dimensional ordered cluster analysis of component groups in
self-organization. Selforganizology, 1(2): 62-77
Zhang WJ, Schoenly KG. 1999. BOUNDARY: A Program for Detecting Boundaries in Ecological Landscapes.
IRRI Technical Bulletin No.3. International Rice Research Institute, Manila, Philippines

IAEES www.iaees.org