How to calculate the percentage of each color in this image ?

How to calculate the percentage of each color in this image ?

[Cris LaPierre]

Do you need help with how to calculate a percentage, or how to get the numbers from your figure?

If the first, . If the second, please share the code you used to create the figure.

[mohab hamed]

i don`t create this image it is orginal

[Cris LaPierre]

Then the first answer is what you need.

[mohab hamed]

I want to calculate the pixels of each color to calculate the percentage of each color .

[Image Analyst]

We went over all this a few days ago.

https://www.mathworks.com/matlabcentral/answers/701477-how-to-convert-pie-chart-into-values-as-shown-in-this-image-by-matlab?s_tid=srchtitle

Was that you (by a different name)? Because the task and images look very similar. Or was that a classmate of yours and this is a class assignment? I showed how to get the colors in the legend, then erasing the legend and getting the percentages in the chart. I know it had pie charts and maybe bar charts too.

[mohab hamed]

He is my classmate, he and I are working on the same task, he told me to ask here because there are many people here to help you

[Image Analyst]

Yeah, that code above is FAR too simplistic to do it. In the other thread, I found the legend box, cropped it out, found the colors in the legend, and then used color segmentation to find the same colors in the rest of the image. Then it found the words next to the colors in the legend and did OCR on them to get the category names.

It's somewhat robust but not bulletproof. In some situations, like really low resolution images it will fail, like you won't be able to find the box around the legend or the characters are too quantized to be uniquely identified by the ocr() function. And if the image did not have a box around the legend, then it will not find the legend. The algorithm was developed for a certain kind of image and if you have another kind of image (say, a gray scale image with hashing patterns instead of colors) then it will fail. You'd have to make the algorithm more sophisticated to detect in advance what it can and cannot work with.

[mohab hamed]

Which function do I use in order to find this legend to crop it ?

[Image Analyst]

If the box is dark and continuous (unbroken) enough then you can simply threshold

boxMask = grayImage < somethreshold;

Now you have to throw away everything outside the box. If everything else touches the border of the image you can do

boxMask = imclearborder(boxMask);

Then fill what's left

% Might need to take the largest blob that's left
boxMask = bwareafilt(boxMask, 1);
% Fill holes to fill in where the letters were
boxMask = imfill(boxMask, 'holes');
props = regionprops(boxMask, 'BoundingBox')
croppedImage = imcrop(rgbImage, props.BoundingBox);

[Adam Danz]

Here's a complete demo showing how to compute the number of pixels per color in an image. That can easily be used to compute the percentage of each color.

https://www.mathworks.com/matlabcentral/answers/592402-how-many-colors-with-their-name-present-in-the-image#answer_493348

Answer 1

[DGM]

I was bored and thought that the image itself was interesting.

% decoding this image will produce a warning
inpict = imread('image.png');

Warning: PNG library warning:
sBIT: invalid

% isolate the bars themselves
% i assume that the image is clean (NOT JPG)
% and that HSV saturation will adequately isolate the bars 
[~,S,~] = rgb2hsv(inpict);

% threshold S
% pick the largest 5 objects (the bars)
mask = bwareafilt(S>0.5,5);

% get object geometry
P = regionprops(mask,'boundingbox');
bb = vertcat(P.BoundingBox);

% things we know from the plot:
% we know the category names
% we know that counts are non-negative integers
% we know the peak count is 11
maxcnt = 11;

% estimate counts from bar height
% i'm using height instead of area, since there's no reason to 
% incorporate error from irrelevant degrees of freedom
cnt = round(bb(:,4)./max(bb(:,4))*maxcnt)

cnt = 5x1

4 8 11 6 5

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

% calculate contribution from each category
% if you want to make these numbers less useful,
% feel free to multiply them by 100.
pct = cnt./sum(cnt)

pct = 5x1

0.1176 0.2353 0.3235 0.1765 0.1471

<mw-icon class=""></mw-icon>

<mw-icon class=""></mw-icon>

The given image is a fully opaque uint8 RGB image expressed as uint8 RGBA with a solid alpha channel, and encoded in PNG with an improper 3-term RGB sBIT specification ([8 8 8]) instead of the required 4-term RGBA sBIT specification ([8 8 8 8]).

In other words, this file was written as RGBA with RGB sBIT spec, when it should have been written as either:

• RGB with RGB sBIT spec
• RGBA with RGBA sBIT spec

I have no idea what software created this bogus file, but the metadata just says 'Google'. Either the encoder is improperly setting this field with default values, or the user is explicitly specifying them incorrectly. The fact that they're native 8b/ch suggests that a user would have no reason to set them anyway.

Note that this given image makes several things reasonably clear. We know that these are discrete counts, so the bar heights represent non-negative integers. It's seemingly clear what the value of the tallest bar is. If that reference value were not obvious, the whole process would be more complicated.

Similarly, the process would be more complicated if saturation were not adequate for isolating the bars from other parts of the image. If the bars were drawn with a very thick border (as in the prior pie charts), the error would be increased accordingly. While integer rounding may be adequate to suppress such a problem in this specific case, we can't always rely on the represented values being small integers.

Likewise, we're assuming that the tallest bar is not actually truncated at the plot box extents. That's not always an accurate assumption.