|M.Sc Student||Amiram Allouche|
|Subject||Enhancing Resolution by Introducing Motion|
|Department||Department of Electrical Engineering||Supervisor||Professor Emeritus Feuer Arie|
It is well known that, when an image (signal) has a bandwidth which is in some sense large compared to the sampling rate of sensor used to digitized it, aliasing errors occur. The limiting rate is called the Nyquist rate. On the other hand, it has been shown in several published results in recent years that, one can use a number of different images (frames) of the same scene to generate an image of improved resolution. This approach, commonly referred to as “Super-Resolution” (SR) or resolution enhancement, has drown significant attention among researchers in recent years.
In this work we are looking at the following problem:
Given a digital image acquisition device, how, and under what conditions can we retrieve an image with truly higher resolution then the device’s sensor limit.
The idea we investigate is to add motion to the sensor and sample in time as well. This way a set of frames is created. Since we assume the only temporal changes are due to the sensor motion, the motion is global.
In our model, we show that the set of M shifted frames generated by motion with temporal sampling, is equivalent to a single image, sampled at M times the original sampling rate, but with recurrent (non uniform) sampling.
By utilizing Papoulis GSE and its multi dimensional extensions we examined the conditions for reconstruction from recurrent sampling generated by the motion.
We consider four basic types of motions: Constant velocity, constant acceleration and two types of periodic motions. While the first two motions were previously treated in the literature, using a different method known as “Motion compensated filtering”, to the best of are knowledge, periodic motions are considered here for the first time.