Problem with creating .bmp images

Hi, all

I'm running PCA on ORL human face image database. What I want to achieve it to extract the principal components, then add the mean vector back to these component. By writing these components as .bmp images, I can get eigen faces. Here is my code

clc
clear all
close all

path = 'D:\\Matlab 7.0\\work\\ORL\\';

data = zeros(400, 10304);

for i = 1:400
    name = [path 'face' num2str(i) '.bmp'];
    temp = imread(name);
    temp = reshape(temp, 1, 10304);
    data(i, :) = temp;
end

meanface = mean(data, 1);

data = data - repmat(meanface, 400, 1);

intermat = (data * data') / (400-1);
[intervec, val] = eig(intermat);

pc = data'*intervec;
pc = pc';

for i = 1:400
    pc(i,:) = pc(i,:) / sqrt(dot(pc(i,:), pc(i,:)));
    sd(i) = sqrt(val(i,i));
    eigenvalue(i) = val(i,i);
end

pc = flipud(pc);
sd = fliplr(sd);
eigenvalue = fliplr(eigenvalue);

%%%%%%Write images %%%%%%%
for i = 1:399
    mode = pc(i,:) + meanface;
    mode = round(mode);
    mode = reshape(mode, 112, 92);
    name = ['pc' num2str(i) '.bmp'];
    imwrite(mode, name, 'bmp');
end

I was expecting to see ghost-like images. However, this piece of code writes images that all white.

May I know if there is anything wrong with my code?

Thank you very much for your kindly help.

Best regards Wenlong

Answer 1

[Image Analyst]

I don't believe BMP supports floating point. If you want floating point, save as .mat. If you want a standard BMP, then cast to uint8 before saving, though beware that you may lose numerical resolution as it will truncate the values and may require scaling and shifting into the 0-255 value range.