| Περίληψη: | Super-Resolution (SR) are techniques that produce high-resolution (HR) images from several observed low-resolution images (LR) images, thus increasing the high-frequency components and removing degradations introduced by the image acquisition process of the low-resolution camera. The main concept of SR techniques is to combine the non-redundant information contained in multiple low-resolution frames to produce a high-resolution image. Single-image interpolation is a closely related technique with SR, which can also increase the image size. However, the lack of additional information provided, leads to limitation of the single-image interpolation quality due to the ill-posed nature of the problem and thus, the lost frequency components cannot be recovered. On the other hand, in SR methods additional low-resolution images are available for reconstruction, making the problem better constrained. The non-redundant information contained in the multiple LR images is introduced by sub-pixel shifts between them. The presence of these sub-pixel shifts may be caused due to motion between the image acquisition system and the observed scene (known as global motion), or movement of objects. The proposed Superresolution algorithm consists of three main steps: sub-pixel shift estimation, interpolation and image restoration. The results of each step are demonstrated and analyzed and it is concluded that more accurate sub-pixel shift estimation can lead to even better results of the Superresolution method.
|
|---|
