Dear Mr.  Karthikumar Sankar,

it is better to program in C or any other high level programming language. In addition, you should provide and example of the image and indicate what features you want to extract. I hope my suggestion will help you.

Dear Sankar

hi

I'm so appreciated if I can help you to make a sufficient GUI software for your purpose.

please let me know what you need and what your images are.

Dear Karthikumar,

Your idea is very nice. And I think MATLAB may be a good tool because it has an image processing toolbox that has ready made functions for image analysis. If I assume that you are doing a research, I discourage using other high-level programming languages, like C (as suggested by one researcher above), because they may lengthen your development time, particularly if you are new to such languages.

Now, coming back to your question, i think it is a bit general. We would like to know the following points:

(1) What features you are interested to extract

(2) What is the nature of the images, and the imaging conditions they are subjected to

(3) What level of accuracy you want features extracted

(4) How you want to extract the features

Given the answers to these problems, we can try providing some guidance. There are tons of algorithms to extract features, each with different characteristics (accuracy, robustness, stability, and computational efficiency); selecting one, therefore, require that the problem be clear and well formulated.

Why not using ImageJ? or Rapidminer?

Hi Karthikumar,

As suggested by others, it would help if you post an example image or two. Depending on the kind of image and the features you want to extract, you might get helpful code snippets as answers here, or be directed towards existing code elsewhere.

It would be beneficial if you already have some exposure to image processing and programming (especially to MATLAB since that is what you want to work with). Otherwise, you might have to spend time getting a handle on the basics.

If you are not already familiar with MATLAB, you might want to check out Fiji (http://fiji.sc/Downloads) which is a version of ImageJ. It is a widely-used image processing tool in the biological research community. You can interactively experiment using your test images and develop a sequence of steps which work for you.

Thank you all for your support and suggestions.

Herewith I have attached the sample microscopic images for your reference.

A transparent polymer sheet was coated with gel at different conditions. Binding of the gel on the polymer sheet varied. At some conditions, the gel slowly washout from the sheet. The images are the microscopic image (1000X) of the polymer sheet where the gel was washed out at different level. The images are taken in a light microscope using Digital camera ( 12 megapixel ). Later selected region from the microscopic images is cropped into uniform size in Adobe Photoshop. From these, I want to know the amount of gel which is still bound on the polymer sheet. According to my knowledge on Image processing, the area of gel (Black regions in the images) and visible polymer sheet (white regions in the images) help to calculate. If any other features, please comment. That will help me to discuss more. Further the images are not taken purposefully for image processing. I want to do the analysis with available images. If there any specifications for images to analysis please comment. I capture again and share. Overall I am looking a Matlab code to calculate the black and white area from the images.

Hi

What I think you must do is the following (in Matlab):

1-Transform the image in a 2D array of numbers, where the higher the number the darker is the pixel.

1-Determine the background level (the value of the pixels in the white area). Initially you could do a visual inspection of the matrix an introduce the value.

3-You create loops to scan line after line, reading the numbers and comparing them to your background. You would have to introduce as well a threshold above which the pixel value is considered to represent gel. You count the number of pixels above the threshold. You have the total number of pixels. Therefore you will be able to compare both numbers and obtain -for instance- which percentage of the area is covered by gel.

Hello Karthikumar,

Despite a good problem description you have provided, as well as the source images included in the post, I have one more question. You have said something about computing the amount of gel embedded in the polymer sheet; (I am not expert in Biology) does this amount mean NUMBER of gel items, like 270 or 700, or covered area? But, from the info, this is called an image segmentation or feature extraction problem. We need to segment (categorize) that input image into two regions (back and white), then compute the area (if "amount" means area) of the black or white regions. The total area is known--the size of your image. I have included a sample segmentation result for one of your images, but that just detect edges, so we need another "filling-operator" to fill the holes, and MATLAB has those tools.

I think here are the necessary steps:

(1) Convert the image into grayscale, so you can easily deal with the problem

(2) Select an edge detector, and segment your image into the interested regions. Few examples of classical edge detectors ere Canny, Sobel, Prewitt,......

(3) Fill the regions enclosed by boundaries

(4) Select the algorithms for computing the area of a region in the image; such algorithms are many online.

With the filling operation--third step in my previous post--I refer to the image attached below(this is the corresponding filling operation of the previous image, but not high performance; just to show concept).

With the binary image, it becomes less challenging to compute size or area of an object. Cleaning of the unwanted artifacts can be done using some morphological operators in image processing.

(I am referring to my previous posts) Then, you can develop a thresholding policy to count the "blacks" and "whites" regions: If, for example, you say all values above 150 (whites) are 1's and the rest are 0's (blacks), you can obtain the percentages of these regions; say "u" is your m*n thresholded image image (containing only 1's and 0's), percentage equals (1/(m*n))*sum(u(:)). But, there may be some other more accurate methods I suppose.

Thank you Prof.Baraka Maiseli. I want to know the total area covered by the gel on polymer sheet. If I know the area then I can calculate the amount, i.e. volume of the gel embedded on the sheet. I try your suggestions and let you know.

Dear prof.Baraka Maiseli,

This is one of the the Matlab code for sob-gradient which I received during one of the Matlab workshop. I tried that for my image and the result is posted here.

Sob- gradient

clc;

clear all;

close all;

cd Images

[file,path] = uigetfile('*.*','Pick an Image File');

a = imread(file);

cd ..

a = imresize(a,[256,256]);

if size(a,3)>1 %%%%%Grayscale Conversion

a = rgb2gray(a);

end

figure;

imshow(a);

m = [-1 0 1;-2 0 2;-1 0 1]; %%%%sobel operator

x = conv2(a,m','same'); %%%%%performing convolution

y = conv2(a,m,'same');

gr = sqrt(x.^2 + y.^2); %%%gradient magnitude

figure;

imshow(gr,[]);

title('Gradient of input image');

imcontour

After this I don’t know how to find out the area of the selected regions. Please help in this.

Dear Prof.Baraka Maiseli,

This is another Matlab code for Kmean-segmentation which I received during the same Matlab workshop. I tried this also for my image and the result is posted here.

Kmean-Segmentation code

clc;

clear all;

close all;

cd Images

[file,path] = uigetfile('*.*','Pick an Image File');

Input = imread(file);

cd ..

Input = imresize(Input,[256,256]);

if size(Input,3)>1 %%%%%Grayscale Conversion

Input = rgb2gray(Input);

end

figure; imshow(Input);

[r c] = size(Input);

Clusters= 4;

Length = r*c;

wd1=r; wd2=c;

%%%%%%% 2D into 1D image

data = reshape(Input,[Length,1]);

Cluster1=zeros(Length,1);

Cluster2=zeros(Length,1);

Cluster3=zeros(Length,1);

Cluster4=zeros(Length,1);

Dataset=data;

%%%%% Initialize the centroid values for each cluster

Imin = min(min(Input));

Imax = max(max(Input));

range = Imax - Imin;

stepv = range/Clusters;

incrval = stepv;

for i = 1:Clusters

T(i) = incrval;

incrval = incrval + stepv;

end

%%%%%% distance between centroids and intensity values

for i=1:Length

for j = 1:Clusters

temp = Dataset(i);

difference(j) = abs(temp-T(j));

end

[y,ind]=min(difference);

if ind==1

Cluster1(i) =temp;

end

if ind==2

Cluster2(i) =temp;

end

if ind==3

Cluster3(i) =temp;

end

if ind==4

Cluster4(i) =temp;

end

end

%%%%% 1D into 2D image

AA1=im2bw(reshape(Cluster1,[wd1 wd2]));

AA2=im2bw(reshape(Cluster2,[wd1 wd2]));

AA3=im2bw(reshape(Cluster3,[wd1 wd2]));

AA4=im2bw(reshape(Cluster4,[wd1 wd2]));

figure('Name','Km Clustering Results','Menubar','none');

subplot(2,2,1);imshow(AA1,[]);

title('Cluster1');

subplot(2,2,2);imshow(AA2,[]);

title('Cluster2');

subplot(2,2,3);imshow(AA3,[]);

title('Cluster3');

subplot(2,2,4);

imshow(AA4,[]);

title('Cluster4');

I don’t know what to do further…

I am seeing that the edge-detection and region-detection results in all cases are good. The supporting codes are also okay, but i would rather recommend that you choose ready-made edge detection functions in MATLAB. For example, the function "J=edge(I,'sobel')" would return edges of an image I using the "sobel" operator, although other options are available (canny, prewitt, etc). That is just one line of code.

I suggest that in the first stage you select the most appropriate edge detector to work with. The criteria for selecting one may be:

(1) Algorithm that captures edges well

(2) Algorithm that is robust and has better capabilities to reject false edges

(3) Algorithm that can identify real edges even when images are degraded by noise or other artifacts.

Then, as i proposed in my last post, you fill the holes or regions bounded by the edges; but care should be taken in bounded regions that contain some sub-regions of the same class (like the first image in your last post, where the Sobel operator was used to detect edges). In MATLAB, the function "imfill(I,'holes')" can be used to fill holes. Then you will have black and white filled-region parts.

To estimate the area, there are also MATLAB functions for that purpose. But, you may try a simple operation I introduced before:

Area_white = (1/(M*N))*(sum(U(:)))*100.

Of course that will just give you a rough estimate. I have tried that simple formula for the image above and got that 16% of the image is covered by that black stuff.

I get 69% white portion in the attached image.

However, I am seeing some challenges in the second and (maybe) the first image because the colors are not well balanced. Therefore, before doing the segmentation you may need to equalize colors; in MATLAB, just one line of code--using the function "histeq", like J=histeq(I)--and you may select the option for adaptive histogram equalization.

If you perform Histogram Equalization in the second image (using the above function) and threshold the image you will get the attached image. Just one line of code. And you can determine the area as before.

The above image has about 50% white portion. Try for the remaining image. Note that you don't have to write many lines of codes, maybe two or three.

This is for the first image.

As a future consideration, you can use the MATLAB coder feature and fully convert the code into some high level programming language, like C or C++, then embed the algorithms into an embedded hardware. This way, you automate the entire process and eliminate manual operations.

Dear Sir, we can do extract basic features from the image and it will give improved interpretation from the final image

You are right; there are several approaches to address the problem. You may select the one that gives useful results. As you understand the (Biology) subject better, you possibly know the basic features you want to extract. In image processing, we are interested to understand the characteristics of those features. Can you provide those characteristics from the sample images you have, apart from "blacks" and "whites" regions?

About interpretation of the final image, my opinion is that we are not very much concerned on the visual appeal or appearance of the final images; Of course, I agree with you that the processed images (posted in my previous posts) may make little sense, particularly for a non-expert in image processing. But, our goal is to get the area using some image processing tools. The input images have been converted into a form that is much easier to work with--binary images. The Region Of Interests (ROIs) have been extracted, and the final images contain only those regions (1's and 0's, 1 - White, 0 - Black). Unless you have other intermediate states apart from "whites" and "blacks" that you want to extract, I think the previous descriptions may help to address the problem.

Try to compare the original and segmented images in the attached images below, the observe if the interpretation of the processed images make sense.

From the above images you the end users can evaluate the results of the program. Otherwise you can mention your problem after observing the final image

Compare the images (original and processed) by analyzing the patterns of their contents.

Dear Prof.Baraka Maiseli,

This is what i am expecting. From your last post i can see the observation better. Is there any option to display the total area, black area and white area at the unit of cm2

I think in that case we have to convert Pixel count into cm. Because 1pixel = 0.026458333cm, then if we have Area_pixel = 50 pixel^2, then we have Area_cm = 50*(0.026458333)^2;

The above calculations are estimates, and I assume that pixel count is proportional to area; but you may try to find other approaches to calculate the area of irregular shapes in the image. The biological process you have generates non-trivial shapes which may need some rigorous maths to calculate their areas.

But the area we are talking about is that on the image, right? And which does not reflect the actual areas on the real sheet, right? Then, I think using ratio or percentage may be a good way to present your results, as that will perhaps be constant on the image as well as on the real event.