Andreea Barac
Department of Information and Computing Sciences, Utrecht University
Nowadays, a lot of captured signals represent a mixture of two or more original signals and the necessity of automatically separating them
into original sources arises. The separation of a set of signals from a set of mixed signals without the aid of additional information is
called blind source separation. I dedicated my experimentation project (under the supervision of dr. Robby T. Tan)
to study this problem for the case of image mixtures formed when taking
a picture of a reflective surface (i.e, painting protected by glass). A lot of researchers addressed this problem and proposed different methods.
Some of them are variants of some popular methods in blind source separation, like Independent Component Analysis, others are based on simple
image properties or statistics. I analysed different existing methods for image separation and focuses on the analysis and
experimentation of an algorithm developed by Gai et al, Blind Separation of Superimposed Moving Images using Image Statistics. The
algorithm assumes that the mixtures are linear, with unknown linear mixing coefficients and unknown motions of sources in each image and it is
based on the statistics of natural images. Besides the separation of the original sources, the method can automatically identify the number of
original images and it has good results even in under-determined cases, where mixtures are fewer than layers. By experimenting the method, I
identified some small drawbacks and gave their possible explanation. Even if this method has impressive results, it doesn't work in real time;
thus, there is still a lot of room for improvement in the field.
Blind Separation of Superimposed Moving Images [PDF], Andreea Barac

|
Input |
Output |
|
|
|
|
|
|