can I take pca analysis of columns extracted from pixel values of a converted to gray image?

starting columns and ending columns

I made a code that outputs the pixel values of the columns from an image.

meanpixelvalues(i) = meanpixelvalues(binaryImage(masks));

I then applied pca analysis to the pixel values and the pca code output 1:

coeff = pca(meanpixelvalues(i))

Results:

coeff =

1

but I don't think that's physically possible and i'm wondering what did i do wrong to get a value of 1 and what does the columns of an image mean?

Thank you.

[Chris]

What is the size and what are the values of meanpixelvalues(i) when you get 1 from the pca?

[Neo]

Going to send it to shortly I have to rerun the code

[Neo]

@Chris

StartingColumns =

Columns 1 through 7

1 435 869 1303 1737 2171 2605

Columns 8 through 13

3039 3473 3907 4341 4775 5209

endingColumns =

Columns 1 through 7

434 868 1302 1736 2170 2604 3038

Columns 8 through 12

3472 3906 4340 4774 5208

[Neo]

rows =

1452

columns =

5208

@Chris

Answer 1

[Image Analyst]

You can do it. I attach an example where I did it for an RGB image where each column was one of the color channels converted into a column vector. But if your variables (features) are individual channels of a gray scale image, like you said, it's difficult to interpret. Why do you want to do it if you don't know what the PCs would represent?

[Neo]

1. So should I convert it back to RGB before doing pca analysis?

2. I know that pca would classify the data set so for it to give me a coefficient 1, wouldn’t make sense, because I’m expecting at least 6 areas of classification. Knowing how it’s classified and into which areas would be sufficient for me.

[Neo]

@Image Analyst

Also,

are you saying that you have to write in the code what you want your pca elements to be? because from your code you have already defined what the PC are rather than telling the code to identify the patters on its own and then you observe what the code tells you are the principal components.

[Image Analyst]

"1. So should I convert it back to RGB before doing pca analysis?" <= no, of course not. That was just an example of how you could use PCA with images. According to what you said, you want the columns (independent variables/predictors) to get the columns of the gray scale image. And we talked about how we don't know how to interpret the results. Perhaps if I saw an example of the image I might know if PCA makes any sense for it or not.

No, you do not tell it what the PCA elements are. You give it your input data, and then it gives you PCs as an output.

I don't know what 6 classifications you want. Is it some kind of spatial grouping of regions with similar gray levels, or some sort of grouping of the histogram (like your histogram looks like 6 Gaussians)?

[Neo]

@Image Analyst

i don't want 6 nor do I have an idea of how many there will be, that is an example number but I know that I want a smaller number than what i have initially.

What determines if an image will have pc or not? but my image definitley will have pc its just extracting them is the problem. and outputting coeff 1 is less than ideal and unrealistic. What i want are DATA CLUSSTERS FROM THE IMAGE THAT I can plot. image attached.

[Image Analyst]

In my opinion, PCA would not be used on that image. I just can't see how it would help or what the variables would be.

[Neo]

sorry I attached the wrong image. But I think my main question is how to extract pc from an image that can be plotted on a graph not from color channels.

[Image Analyst]

You have to have 2 or more features. What are your features?