1) After some weeks of work, we have done some great improvements on the project. At the moment, we have already applied all the algorithm of the fingerprint in Matlab and we can say it is working more or less. In order to test the quality of our fingerprints, we have made several tests adding noise to the video. The fingerprint detects correctly the film, when we add Gaussian and random noise to the input video. It works perfectly when we multiply with a scalar the video. But it doesn't work well when we record the scene with the cellphone. We recorded the video of Full Metal Jacket with our phone and the program sais it is Blade Runner. However, we have to say that the recording conditions weren't very good. So we can say that the algorithm is working quite well and we are going to continiue in the same direction. In the next week, we are going to make several tests with different types of noises, try to improve our algorithm and deduce conclusions.

 

  2) We are using the Haar wavelet transform, it was mentioned in one lecture. We have been informing and this transform is better when we are processing video. So we decided to follow the paper and use the Haar wavelet transform, instead of the Fast Fourier Transform.

 

  3) A tool we are using in the project that is out of this class, is the minimization of the Least Squared Error. We have created the projection matrices of the fingerprints using Singular Value Descomposition, and then projecting the noisy fingerprints see from which of the film spaces is closer, in order to decide which film is detecting.

 

  4) It has been cool to see that implementing the algorithm as it says in the paper the program works. And we have discovered the function "resize" in Matlab. This function is very useful in the normalization phase of the algorithm. It resizes the matrix to the dimensions we decide. This way, we have all the videos we input of the same size and the we apply the "rgb2gray" function to have the image in grey scales.