ginput to select RGB values of image to hold over when I turn image to grayscale

Im working on a function to where i can use ginput to selct and area on th image and hole those values when i convert it ot grey scale. Having issues with it and index size.

imagein = 'Bird.jpg';
imshow('Bird.jpg');
n = 4;
% read and display color image
A = imread(imagein);
pic=uint8(A);

grayscale(:,:,1) = uint8(0.2989 * pic(:,:,1) + 0.5870 * pic(:,:,2) + ...
    0.1140 * pic(:,:,3));
grayscale(:,:,2) = grayscale(:,:,1);
grayscale(:,:,3) = grayscale(:,:,1);

[x,y] = ginput(2);
% Put a cross where they clicked.
hold on;
plot(y:x)
% Get the location they click on.
row = int32(y);
column = int32(x);
% Extract the RGB color from that location.
selected_color = pic(row, column, :)

gray_color(:,:,1) = selected_color(grayscale(:,:,1)+1);
gray_color(:,:,2) = selected_color(grayscale(:,:,1)+1);
gray_color(:,:,3) = selected_color(grayscale(:,:,1)+1);

subplot(2,2,1)
image(pic)

subplot(2,2,2)
image(grayscale)

subplot(2,2,3)
image(gray_color)

[DGM]

I'm not really sure what exactly you're trying to do with this. You're selecting two points. The way you index when calculating selected color would give you the corners of a rectangle, but I can't imagine that's what you want. You can use getrect() to obtain the location and geometry of a selected region.

Are you trying to sample the colors in a rectangular region and then retain instances those colors throughout the image, or are you simply trying to retain the sample region itself? One of these is much easier than the other.

The latter case being something like this:

[luke schneider]

ill use the peppers image thats preloaded in matlab as an example. if i use ginput and select areas on that image it will have values for where you clicked. now im trying to figure out how to use those values and any other part of the image with the same value while the rest of the image is grayscale. so if i clicked on the red pepper it would hold its value and with simalr value while the rest of the picturewould go grayscale

[DGM]

Okay, that's a more complicated problem. Normally, the way I handle color selection is to get a point, find the average color around that point, and then find all points in the image whose color is within some weighted euclidean distance from the sample average. In this case, I'm only finding distance from one color to many colors.

What you're doing is sampling an abundance of colors in a rectangular region and then comparing many colors to many colors. It's unclear whether you want to test for exact color tuple matches (which may be technically useful, but visually unsatisfactory) or if you want to test based on color distance.

I can make an example for those requirements (rectangular sample region, exact matching), but if the requirements are flexible, there are color-based mask generation UIs available.

[luke schneider]

yes i was looking for exact color tuple matches. Which would probably be better to convert the image to VHS first righ?

[DGM]

For exact matches, the model isn't going to matter. Let me throw together a demo of what you're talking about and we'll see.

[DGM]

This is the result from testing every color in the selected region for exact matches throughout the image:

on the other hand, doing a crude distance matching using a number of individual selected points in the same region gives this:

The second case was done with a gui tool and can likely be improved if object selection is intended.

[luke schneider]

The second case would be what im trying to accomplish

[DGM]

That second case was an ellipsoidal (euclidean distance) match in LCHab. I simply did that by hand using immask() from the MIMT.

https://www.mathworks.com/matlabcentral/fileexchange/53786-image-manipulation-toolbox?s_tid=srchtitle

If you're needing something to build into your own GUI or something, maybe you can use it or ImageAnalyst's demo as an example.

Once you generate a mask, the rest is just a matter of applying it to make a composite image:

gray_color = pic.*mask + grayscale.*(1-mask);

or for ML versions older than R2016b:

gray_color = bsxfun(@times,pic,mask) + bsxfun(@times,grayscale,1-mask);

bear in mind that you may have to cast the mask to match if pic or grayscale are integer class.

[luke schneider]

greatly appreciated

[DGM]

I should point out that there are other color-based mask generation tools around. If you have IPT, you can try the Color Thresholder app.

Answer 1

[Image Analyst]

The red pepper has thousands of red colors in it. Do you want to have only the two pixels you clicked on stay red and all other pixels turn gray? Or do you want to find all colors within a certain delta E color distance of the red you clicked on stay red and all other pixels turn gray? See attached demo.

[luke schneider]

Correct i wanted all colors within delta E color that i click on stay the same while other pixels turn gray

[Image Analyst]

So this is what the demo does now. Can you modify it to make it gray outside the mask? Not too hard but let me know if you can't figure it out.