How to make a similarity analysis of several colormap plots

I have one colormap plot having red, blue and white colors in it. This is my reference figure. Besides, I have several output figures having the same scale and dimensions. I would like to make a comparison of the outputs with the reference figure and give scores to each output to measure the prediction capacities of them. I will ignore the white colored areas in the reference. For the blue and red colored areas, if the output has the same color at the same location with the reference, it will be a plus to the score. If it does not predict the color at the exact location there will be no penalty. I want total scores of each output to make comparison. Can anyone help me to apply this?

[Adam Danz]

There are many types of plots that use a colormap and color is set differently for different types of plots.

What you want to compare isn't the color, it's the values that define the colors but since we don't know what function and syntax you're using, that's about as far as I can go with details.

[Burhan Yildiz]

I am comparing measured and modeled topographic maps. The reference is obtained from the satellite images before. I do not totally know the exact steps in developing this. The outputs are generated in a matlab based tool 'quickplot' using a colormap selection of jet.. I am not sure if this is what you meant.

[Adam Danz]

What's quickplot???

Are you talking about the software quickplot that is independent from Matlab but can read [some] matlab files?

[Burhan Yildiz]

No, that is another software with the same name. I am talking about this one Manual.

[Adam Danz]

I'm not familiar with that software but I see that its created using Matlab.

In any case, my first comment still stands. Color is merely a depiction of some set of values. If you want to compare two sets of colors on the same scale, you need to extract the values that gave rise to those colors.

You mentioned a "colormap plot". "Colormap" in Matlab is what defines the range of colors for some dataset. Colormaps can be applied to plots (examples: pcolor, contour, surface, etc). We don't know what type of plot you're working with or what variable defines the color of each element.

[Burhan Yildiz]

I see your point related with the colors and the values. I have the values for the model outputs but I don't have the values for the reference figure that is previously obtained by satellite images. So that I want to make the similarity analysis based on colors.

About your comment related with the colormaps, I am using this software with a gui and I am using the terminology there. I am using a flag type of colormap. Simply, in a 2D region, this option turns the cells having values greater than 0.5 to red, smaller than -0.5 to blue and between -0.5 to 0.5 to white. I have the color definitions of red, blue and white colors and I can share but I don't know whether it is relevant or not. I am attaching one sample output. Hope it will help. I have one more plot with the same scale and I want to apply the similarity analysis that I explained above over two of them.

[Burhan Yildiz]

By the way the color scale in the sample is not compatible with what I wrote above. I just noticed and you may ignore that also.

[Adam Danz]

If you can get the handle to the figure (try gca) then you can probably get the handle to the objects plotted on the figure. Using the object handle, you can get the zData which is probably defining the colors.

[Burhan Yildiz]

I dont have access to the original figure data.

Answer 1

[yanqi liu]

use color hist to compute

or

you can use sim hash functions

[Burhan Yildiz]

Can you elaborate how to use these methods?

[yanqi liu]

such as use hist to make target follows

clc; clear all; close all;
I=imread('01.jpg');
rect_init = [198 208 41 27];
[It,rect]=imcrop(I, rect_init);
[rows,cols,chs]=size(It);
y(1)=rows/2;
y(2)=cols/2;
tic_x=rect(1)+rect(3)/2;
tic_y=rect(2)+rect(4)/2;
md=zeros(rows,cols);
h=y(1)^2+y(2)^2 ;
for i=1:rows
    for j=1:cols
        dist=(i-y(1))^2+(j-y(2))^2;
        md(i,j)=1-dist/h;
    end
end
cm=1/sum(sum(md));

hist1=zeros(1,4096);
for i=1:rows
    for j=1:cols
        d_r=fix(double(It(i,j,1))/16);
        d_g=fix(double(It(i,j,2))/16);
        d_b=fix(double(It(i,j,3))/16);
        q_temp=d_r*256+d_g*16+d_b;
        hist1(q_temp+1)= hist1(q_temp+1)+md(i,j);
    end
end
hist1=hist1*cm;

I2=imread('05.jpg');
dif=0;
neg=[2,2];
while((neg(1)^2+neg(2)^2>0.5)&&dif<20)  
    dif=dif+1;
    It2=imcrop(I2,rect);
    hist2=zeros(1,4096);
    for i=1:rows
        for j=1:cols
            d_r=fix(double(It2(i,j,1))/16);
            d_g=fix(double(It2(i,j,2))/16);
            d_b=fix(double(It2(i,j,3))/16);
            d_t(i,j)=d_r*256+d_g*16+d_b;
            hist2(d_t(i,j)+1)= hist2(d_t(i,j)+1)+md(i,j);
        end
    end
    hist2=hist2*cm;

    w=zeros(1,4096);
    for i=1:4096
        if(hist2(i)~=0) 
            w(i)=sqrt(hist1(i)/hist2(i));
        else
            w(i)=0;
        end
    end   
    
    sum_w=0;
    xw=[0,0];
    for i=1:rows
        for j=1:cols
            sum_w=sum_w+w(uint32(d_t(i,j))+1);
            xw=xw+w(uint32(d_t(i,j))+1)*[i-y(1)-0.5,j-y(2)-0.5];
        end
    end
    
    neg=xw/sum_w;
    rect(1)=rect(1)+neg(2);
    rect(2)=rect(2)+neg(1);
end

figure(1);
imshow(I);
hold on; rectangle('Position', rect_init, 'EdgeColor', 'm', 'LineWidth', 2);
figure(2);
imshow(I2);
hold on; rectangle('Position', rect, 'EdgeColor', 'm', 'LineWidth', 2);
figure(3);
plot(hist1, 'r-')
hold on;
plot(hist2, 'g-')

you can see that the color hist can be match

this is meanshift kernel

sir, may be can reference resources, thanks a lot

参考：《计算机视觉与深度学习实战》

[yanqi liu]

01

05

[Burhan Yildiz]

Thank you for the answer. I could not understand two things. What is that rectangle for and how did you obtain the coordinates for it? And, what should I understand from the histogram output?

[yanqi liu]

may be use color hist to make feature vector，and use the rect image to match