Search for: All records

Image processing is an important technique that is used in many fields, such as self-driving vehicles or facial recognition. One method is called image convolution, which involves many calculations that manipulate the pixels of an image to produce a new image with a desired effect. This is computation intensive and requires a significant amount of time when run on a traditional computer processing unit (CPU). Since image processing is used for real-time applications, such as those mentioned above, it is essential that convolution algorithms run as quickly as possible. A common way to speed up image convolution algorithms is to take advantage of the highly parallel structure of graphical processing units (GPU) to perform concurrent calculations. One problem with GPU applications is that they are often limited by the latency delays associated with transferring data between the CPU and the GPU. Previous works have looked into different ways to address this issue and optimize GPU programs. This research aims to explore different memory implementations and compare them to see which is best at optimizing data transfers.