franciscovargas / ivr Goto Github PK

create a new matlab file exactly the same as thresh.m only that the thresholds are not hardcoded but calculated on on the go (for each color channel using findnorm.m).

inspect what difference it makes to threshold in hsv as opposed to rgb

(r',g',b') = (r/(r+g+b),g/(r+g+b),b/(r+g+b)) where r',g','b are normalized. You must construct or find an algorithm which performs this on an image.

After colour normalization has been carried out, histograms (and smoothed histograms, i.e convolved with 1D Gaussian) must be produced and saved as calibration tools for threshold. Here is a small snippet showing how this could be approached:

> [hegiht width] = size(normalized_image);
> red_vec = reshape(normalized_image(:,:,1), 1, height*width);
> green_vec = reshape(normalized_image(:,1,:), 1, height*width);
> blue_vec = reshape(normalized_image(:,:,1), 1, height*width);
> red = dohist(red_vec,1);
> green = dohist(green_vec,1);
> blue = dohist(blue_vec,1);
% create 1D gaussian mask and convolve
> gauss = fspecial('gaussian',[1,5],6);
> smooth_hist = conv(gauss, red);
>  plot(smooth_hist);

Test the new changes (gray scale autocalibration and cropping) in all images.

Must take new pictures in the lab for the assigned surfaces in order to test:
-Lighting
-Positioning and angling of capture
-Surfaces