How to detect a specific image object and the crop object from the input images

Hi, I have solar panel cell images taken from various resolution cameras, I want to detect a cell in the input Image and find out that specific object(cell) inside the input image, How i can do this?? I need to crop cell once it is detected, here is an example, I need to separate out cells based on black rectangle around cell.

Answer 1

[Image Analyst]

Muhammad:

Attached is a full demo that works with your image that you posted. It assumes that the band height is fixed - all tiles within one horizontal band seem to be within the same pair of row numbers. Then it gets the average horizontal profile and looks for little dark lines that separate the 10 different tiles. Then it goes through the band cropping out each tile and writing it to disk. Below is a screen shot after it has found the last tile in band #6, but the screen will update at every single tile. There are tons of comments and helpful pictures to show you what it's doing. Let me know if this works for you. If it doesn't work for some other image, post the image it fails with so we can determine why its different than the only image you posted so far.

[Muhammad Irfan]

I tested this code, It workd for me for the given image, It does not work for me if i changed the image taken by different camera, here is sample.

[Image Analyst]

Chances are the vertical image size will be different with a different camera. Did you try the obvious fix of changing where the lines defining the band separations are? You might have to make an algorithm for that. First I'd get rid of the white letters by finding the mean of the image and setting all white pixels to the mean. Then I'd get the average vertical profile with mean(grayImage, 2) and then find the half max values to identify where the black border starts and stops. Then I'd divide the distance between those by 6 to get the new dividing lines. Try it.

[Image Analyst]

I guess you're having trouble with the algorithm, so I've attached a function that seems to work pretty well.

[Muhammad Irfan]

This really works nice with me....Thanks alot for your time.

[Muhammad Irfan]

One more thing i want to ask, The rows can be different in a pannel. Every time they are not fixed size of row and number of rows can be different. In function "GetRowDivisions". how i can find total rows in image and their dimension?

rowDivisions = [1, 605, 1153, 1701, 2252, 2794, 3348]; % Initialize

[Image Analyst]

You could ask the user with inputdlg() how many rows are in this image, or else you could look at the average vertical profile, then smooth it and subtract it to find the peaks - much like I already did to find the divisions horizontally. The code is already in there, just copy it and adapt it to work on the vertical profile instead of the horizontal profile.

[Muhammad Irfan]

There are too many peaks in the difference signal in case of vertical profiles because of lines within the cells. Can you suggest any method through which I can get only the edge peaks for each cell ??

Answer 2

[Image Analyst]

Don't you have these 60 images "taken from various resolution cameras"?

Or is all you have one gigantic image where someone unfortunately stitched them all together? And if so, why can't you just use imcrop() to chop the image into 10 columns and 6 rows?

[Muhammad Irfan]

No this is single image I have a dataset of 9K+ images.I need to write a script so that I can separate all cells from 9000+ images.Using imcrop it will be hard.I need to detect images in very fine manner.Some images have very thin edges to determine uniqueness of one cell from other.

Answer 3

[Matt J]

My strategy would be as follows. I demonstrate with a somewhat cropped and binned version (attached) of your original image. First do some enhancement of the image to emphasize the horizontal and vertical lines over the background,

A=imread('Panels.png');

A=medfilt2(A,[10,10])-A;
A([1,end],:)=max(A(:));
A(:,[1,end])=max(A(:));

%%enhance columns
B=A;  d=ones(10,1); 
B=imerode(B,d);

%%enhance rows
C=A;
C=imerode(C,d.');

D=(B+C);  %combine

Then, because you know the approximate dimensions of the cells, you can pick out a box enclosing the first cell

OneCell=D(1:130,1:130);

figure(1);
imagesc(OneCell)

Then you take row-wise and column-wise means of this cell to generate 1D profiles. The spikes in these profiles tell you the row/column locations of the horizontal/vertical lines, from which you can determine the boundaries of the cell more precisely and then crop. Knowing the boundaries of the first cell, you also then know the upper left coordinates of its neighbors and can do similar searches for them, and so on...

figure(2);
subplot(2,1,1);bar(col) 
subplot(2,1,2);bar(row)

Answer 4

[Huzaifa awan]

i want help in same issue

[Matt J]

Then you have it. Several solutions have been posted above and the OP was satisfied by one of them.

[Huzaifa awan]

how to convert this code into windows application and how to save each crop image separately in one folder

[Huzaifa awan]

sir, what is the complete code for your last above answer?

[Matt J]

I never coded the whole thing.

[Huzaifa awan]

i run this code run fine but when i create .exe file it give me error .i only request you sir to help me on this issue

[Image Analyst]

Huzaifa, see the FAQ for possible compiler solutions: https://matlab.wikia.com/wiki/FAQ#My_standalone_executable_won.27t_run_on_the_target_computer._What_can_I_try.3F