الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 120 |  |  |
| | | from | 120 | | |
Abstract
Improving medical images is a critical task that requires a powerful system that is very sensitive to any change or transformation in the image. This is necessary to ensure accurate diagnosis and correct treatment. To achieve this, redundancy must be eliminated through signal compression. and reduce the amount of data that needs to be stored or transmitted. However, this requires a computationally complex compression method, which can be difficult to implement in some applications. In medical imaging, over-sampling is undesirable since it can harm the imaged object as it may impair the existing objects. Because of this, certain types of image acquisition systems might not be appropriate for such applications. So, it is important to use a system that can detect any changes in the image without over-sampling, as this can damage the captured object. Compressed sensing (CS) is an advanced technology that has the potential to revolutionize the images’ acquisition and reconstruct. It is based on the idea that signals can be recovered from much fewer measurements than the number of measurements required by the Nyquist-Shannon sampling theorem. This is possible because signals can often be sparsely represented in some domains, and CS takes advantage of this fact to compress and reconstruct images efficiently. The development of CS has opened a range of possibilities for image acquisition and reconstruction. For example, it can be used to reduce the amount of data that needs to be collected and stored, as well as to reduce the amount of time required to acquire and reconstruct images. It can also be used to improve the quality of images by reducing noise and artifacts. Nonetheless, there are still some challenges that need to be addressed to make CS a viable technology. The most important of these is the design of the sampling matrix, which determines how the IV signal is sampled and compressed. The development of efficient reconstruction methods is also necessary to ensure that the reconstructed images are of high quality. Single image super-resolution (SISR) is a computer vision problem that has recently attracted a lot of attention from researchers. The goal of SISR is to generate a high-resolution (HR) image from its low-resolution (LR) version. This is a difficult problem to solve because it is a many-to-one mapping problem. However, over the last decade, a variety of traditional non-deep learning (DL) based approaches have been developed to tackle this problem. To address this issue, various methods have been proposed to make the DL-based SR models invariant to the geometric transformations. The spatial transformer network (STN) is a powerful tool for achieving spatial invariance in deep learning networks. It is characterized by its ability to merge into existing Convolutional Neural Networks (CNNs) to provide the ability to mitigate the effects of geometric transformations. A suitable transformation can be carried out dynamically for each input sample using the STN network. This is especially useful for tasks such as image recognition, where the input image may be distorted or rotated. To estimate the transformation parameters and resample the input image, the STN is often employed to conduct bilinear interpolation in the pixel domain. However, this may lead to blurred output compared to the original input image. To address this issue, researchers have proposed a robust deep learning-based high-resolution (HR) image recovery framework. This framework is made to take care of the geometric adjustments while also restoring the HR image from its corrupted form. V This thesis proposes a framework that utilizes a combination of convolutional layers and STN layers to perform the geometric corrections. The convolutional layers are used to extract features from the input image, while the STN layers are used to perform the geometric transformations. The output of the STN layers is then fed into a decoder network, which is used to reconstruct the HR image.