How to process the image in parts

I'm working with object detection, but my object of interest is too small. I would like to split my image into parts and process each part of the image automatically.

Here's an example:

My code below:

clear, clc, close all;
load('Detector.mat');

I = imread('fotos/foto (15).jpg');
[bboxes, scores] = detect(detector,I);

for i = 1:length(scores)
   annotation = sprintf('Soja = %.1f',scores(i));
   I = insertObjectAnnotation(I,'rectangle',bboxes(i,:),annotation);
   
end

figure
imshow(I)
title(size(bboxes,1))

'Detector.mat' is also archived.

I would like you to help me.

[Benjamin Thompson]

Images are just matrices, so you can use colon notation to copy a part of the matrix:

>> A = imread('image.bmp');

>> size(A)

ans =

3067 1540 3

>> A1 = A(1:500,1:500,:);

>> figure, imshow(A1)

[Guilherme Franklin]

Benjamin Thompson, you are helping me a lot.

Thanks.

[Guilherme Franklin]

We could keep in touch on whatsapp to talk about matlab, what do you think?

I am Brazilian.

My whatsapp number is: [phone]

Below is my contact link:

https://api.whatsapp.com/send?phone=5538998326580

Answer 1

[Image Analyst]

If your original image fits into memory, I see no reason why to split it into parts. Certainly a 3067 by 1540 by 3 image is not huge by any consideration. I'm not even sure how you would split it into parts if you can't fit it into memory with MATLAB. I guess you'd have to have some other program that would do that.

You know that even if you can't resolve details due to subsampling to fit onto the display, the underlying image will still have all the information, right? So why do you want to split it into parts?

[Guilherme Franklin]

I want to divide it into parts so that it can be closer and find my objects of interest.

[Image Analyst]

That doesn't make any sense. You can always zoom in with the axes toolbar. The problem with chopping your image into tiles is that each tile may now contain only part of a white ball, thus preventing you from getting an accurate measurement or detection of it. It may be a semicircle if it ends up being on the edge of a tile. Now you can't get an accurate diameter of it so you'll have to exclude it, thus getting rid of some of the balls when there is absolutely no reason to do so. What if you did chop it up into tiles? Then you'd have to have a number of axes, and since your original image it too big to fit on screen, you won't be able to see all the individual tiles simultaneously either. So then you're going to have to have some kind of program to bring the desired tile onto the screen and show it. It's just a lot of work for no benefit. Just do your image analysis and if you want to zoom in, use the axes toolbar to zoom and pan.