Problem in creating image database file.

Hi,

I am trying to code face key points detector using Matlab using NN toolbox. Training text file is of form (string path of face image followed by 4 integers-bounding box):

foldername\filename_0001.jpg 84 92 161 169

5000 rows like this.

How do I create mat file of this cropped face images and pass on as input to network?

Answer 1

[Walter Roberson]

http://matlab.wikia.com/wiki/FAQ#How_can_I_process_a_sequence_of_files.3F

You would imread() each, imcrop(), reshape as a vector, add it to a column to a matrix you are building up. The completed matrix would be passed as the inputs for the neural network.

[Sam]

thanks for the reply, I need little more help now.

problem is after cropping i need to resize it to 39x39.

but my target (x,y)face points are give in original image dimensions(250x250).

how to rescale it to proper values after cropping and resizing for NN target i/p for training?

here is my code,

clear all;close all;

load('jpgfiletrain_short.mat');% loading cell imdata{} that contains 4 images

fileID = fopen('trainImageList_short.txt');

% 4 %u's are square bound box (x,y) pair that vary in size, 10 %f's are 5 (x,y) pair of 2 eye

% centers, 1 nose tip and 2 mouth corners...

C = textscan(fileID,'%s %u %u %u %u %f %f %f %f %f %f %f %f %f %f');

fclose(fileID);

% celldisp(C)

for k=1:4

cropdata{k}= imcrop(rgb2gray(imdata{k}),[C{2}(k) C{4}(k) C{3}(k)-C{2}(k) C{5}(k)-C{4}(k)]);

rszdata{k} = imresize(cropdata{k}, [39 39]);

target(:,k)=[C{6}(k) C{7}(k) C{8}(k) C{9}(k) C{10}(k) C{11}(k) C{12}(k) C{13}(k) C{14}(k) C{15}(k)];

end

[Walter Roberson]

Multiply by 39/250