and test it on the image database DATABASE.
The database consists of 5 image classes, each class contains 6 images, one of them a sketch.
They are named
Please use the MATLAB Canny edge detector on the original images, try to find appropriate parameters.
The main task is the implementation of the angular partitioning. Using the 2-dimensional 'find' command ([r,c]=find(im))
together with transformation of cartesian coordinates into polar coordinates (cart2pol) should make that an easy
task. For the rotational invariance of course 'fft' is the function of choice.
To run test on the database, the 'dir' command is quite useful: it reads the directory into a structure.
is 'names', which you then can use (in a loop) to read all images in the DB.
This is a 2 week group assignment. I expect that the original implementation will fail on the given database.
Please come up with your own ideas to improve the algorithm! Do not try to implement all details of the paper, but the main idea
and see where it fails/works. Especially: do not work too much on automatic optimal edge detection, but, if needed, adjust the parameters
for the edge detection by hand. The main question in this task is: can the angular partitioning descriptor reflect visual
properties of images? Is image similarity sufficiently characterized by it?
Good luck !