Grant: Discovery Grants Program - Individual

Development of Digital Signal Processing Techniques and Architectures for Communication and Biomedical Applications The field of digital signal processing (DSP) has experienced explosive growth during the past couple of decades. The DSP techniques have become an integral part of the products and services needed in our daily lives. The research efforts of the applicant in this area have recently led to very concrete results that have been shared with the international scientific community and have also given rise to new ideas and directions for further investigation. The overall objective of the proposed research is to develop efficient algorithms and architectures for reliable processing of images, video, and biomedical signals and their efficient and cost-effective implementations for communication and biomedical applications. Due to the physical limitations of image capturing devices, the quality of the resulting images is lower than that required by the targeted applications. Image restoration is the process of enhancing the quality of the degraded images. Image restoration has a wide range of applications ranging from intelligent vehicles to medical imaging. In the absence of an accurate degradation model, image restoration is a challenging task. In this proposal, we aim at solving this problem through convolutional neural network (CNN) based image super resolution while overcoming many of the challenges associated with this kind of solution. An accurate human recognition is vital for several security-based applications, such as performing online financial transactions and forensic investigations. Minutiae of fingerprints being stable and robust to their impressions is used in most fingerprint recognition systems (AFRS). However, the accuracy of such AFRS is compromised in view of their small numbers, a situation that worsens in partial fingerprints. We propose to investigate the design of a CNN based AFRS using fingerprint pores, in view of their being level-3 qualitative features, immutability, and uniqueness in terms of location coordinates and large numbers. Feature matching between a pair of images is a crucial task in medical applications, such as medical anomaly detection and image guided surgery, where a real-time matching task often needs to be done. We propose to investigate very fast and accurate schemes for this task based on realistic assumptions about medical images. Alzheimer's disease is the most common dementia affecting millions of elderly globally. There being no cure for the disease, the detection and classification of this disease is crucial for the care of the patients. If the disease is detected early, then certain available treatments can help its progression. We propose CNN solutions for the detection and classification of the disease. The direct beneficiary of this research will be the Canadian telecommunication and biomedical industry through the development of new technologies and HQP from this research.