| Περίληψη: | The purpose of this thesis is to investigate the issue of removing camera shake blur from a single image. The setup of the issue under investigation includes a camera taking a picture under conditions that favor a non-sharp result. These conditions may include dim light or camera motion during a shot. The result is a superposition of pixels of the sharp image leading to a blurry image with unclear details. This is because the blur has removed the high frequency components of the taken image. Dealing with such a matter is most important since there are several cases that blurry camera shots contain very important information for research, medical or image matters and the fact is that these shots cannot be retaken. Such cases can be astronomical images, car plate images or images from medical scans and microscopy. This is also a problem in everyday life photos. For example, pictures from friends’ reunion, wedding pictures or family pictures.
As input we use a single image having no other information regarding the scene the people or the objects present in the shot nor do we possess any other information regarding the equipment user or the user. Additionally, it is assumed that the method will be implemented for embedded systems processors and specifically for the ANDROID platform, meaning that a fast and lightweight implementation is in order. To cope with the aforementioned matters, we present from the current literature the most common methods and after a presenting a benchmarking along with the criteria we select the most suitable method. Specifically, this master thesis has the following structure. The second chapter presents the current literature. Afterwards in chapter three the method is analyzed and in chapter four we present the implementation of the method in MATLAB. In chapter five we present the C/C++ implementation that is used for the android implementation within an android application that reads images from the image gallery upon which the algorithm is applied. Finally, in chapter seven the evaluation and benchmarking is presented.
|
|---|
