What is the function of flag?

Hi,I want to find edge abruptness in melanoma but I dont know what is the function of flag in this coding..can u help me

Answer 1

[Image Analyst]

It appears to be some kind of way to get a mask of the background with smoothed borders. I agree with Walter that it's poorly written code and should be rewritten.

[waffle]

can show the rewritted code to me..tq for your help

[Walter Roberson]

Work through the 16 different possibilities of a 2 x 2 matrix having each value being 0 or 1, and calculate the filter2 result for each. Then study the combinations together with the if/elseif section to figure out which combinations are to produce which results. Then write new code that expresses those combinations clearly.

[Image Analyst]

Give a definition for "edge abruptness", because I'm not confident that that code is computing it as it should, but I don't know the formula for it. I'm guessing you have some paper that defines what it is.

[waffle]

This is the definition according to some paper

Answer 2

[Walter Roberson]

To interpret the code, imagine a 2 x 2 matrix of binary values, and that you are focusing your attention on the upper left corner. Then

if(flag==1||flag==2||flag==4||flag==8)         %true if exactly one of the four pixels is set

elseif (flag==3||flag==10||flag==12||flag==5)  %true if two pixels in a row or column are set

elseif (flag==9||flag==6)                      %true if two pixels on a diagonal are set

elseif (flag==11||flag==14||flag==13||flag==7) %true if three pixels are set

and the code has no case for the situation where none of the pixels are set or all of the pixels are set.

You can use filter2 with [1 1;1 1] to count how many of the four pixels are set. To distinguish diagonals, you can filter2 with [1 0; 0 1] and detect result 0 or 2 as indicating a diagonal.

The distance calculation code D(idx)=sqrt((i+0.5-mx)^2+(j+0.5-my)^2) is exactly the same on all of the branches, and calculates the distance to the centroid. That can be done for all of the points simultaneously instead of in a loop, after which the "no points set" and "all points set" cases can be removed before calculating the mean.

The one pixel and three pixel cases lead to the same addition of sqrt(2) to P, so distinct branches are not needed for the two.

You should be able to vectorize all the calculations.

[waffle]

are freeman chain code used in this codes?

[Walter Roberson]

No, freeman chain codes are not used here.

[waffle]

but i still cant understand what flag mean? if(flag==1||flag==2||flag==4||flag==8) how to decide these number or i can state others value and how to get it?.tq

[Walter Roberson]

Imagine a 2 x 2 matrix of binary values, and that you are focusing your attention on the upper left corner. Then

(flag==1||flag==2||flag==4||flag==8)

is true if exactly one of the four pixels is set.

Another way of looking at it is to take a 2 x 2 matrix, M, out of the image, and take

temp = M .* [1 2; 4 8];
result = sum(temp(:))

or, coded another way,

result = M(1,1)*1 + M(1,2) * 2 + M(2,1) * 4 + M(2,2) * 8;

Now, by examining result, you can figure out which of the four pixels was set. If exactly one of the pixels is set, the result would be exactly 1, or exactly 2, or exactly 4, or exactly 8, so by testing for 1, 2, 4, or 8, you can figure out that one pixel was set. If the upper left and bottom right pixel were set, that would be 1 + 8 = 9; if the lower left and upper right pixel were set, that would be 2 + 8 = 10, so testing for 9 or 10 would be the combination of "diagonal pixels are set". And so on.

[waffle]

ohh..i see thank you so much..is this image is divided into 16 or 8?

[Walter Roberson]

The code temporarily divides the image into 2 x 2 overlapping blocks, but that is all handled by imfilter().

[waffle]

idx=0;
P=0;
e1=sqrt(0.5);
e2=1;
e3=2*e1;

What are these values stand for? why there have 3 values and how to decide it?

[waffle]

and i dont understand what is mean by idx? is that kmean clustering

[Walter Roberson]

P is being used as a running total of some kind of local distance measurement, with the possible distance contributions being e2 = 1 (for adjacent pixels in a straight line), or e1 = 1/sqrt(2) (for one isolated pixel set, or for 3 of the four pixels set), or e3 = sqrt(2) (for two diagonal pixels.) The code you show does not use P after it is calculated; perhaps there was more code later on, or perhaps it was being returned from the function for some reason.

idx is used as a running counter of the number of locations whose distance has been calculated and stored. Not all locations have distances stored, because the code does not store a distance if none of the four pixels are set, or if all four pixels are set.

Answer 3

[Rik]

The answer can be found in the documentation (just type doc filter2).

"The filter2 function filters data by taking the 2-D convolution of the input X and the coefficient matrix H rotated 180 degrees. Specifically, filter2(H,X,shape) is equivalent to conv2(X,rot90(H,2),shape)."

So flag is the output of that filtering in some specific pixel.

[waffle]

tq.. so how this flag values determine by (flag==1||flag==2||flag==4||flag==8)? How to state those values?

[Walter Roberson]

It appears to be a hack to test exactly which neighbours are set. I would recommend rewriting the code.

[waffle]

can you show the rewrite code to me?thank you so much

[Walter Roberson]

I described how to rewrite it, above, https://www.mathworks.com/matlabcentral/answers/336529-what-is-the-function-of-flag#answer_264081

[waffle]

thank you so much for your help!