Implementation of various edge detection filters using different thread distributions

A dedicated framework with memory interleaving and parallel handling strategies can lessen the weight of host CPU along these lines making the framework more appropriate for ongoing applications. Presently it is conceivable to use parallelism utilizing multi-cores on CPU however it should be utilized explicitly to gain superior performance. Latest GPUs has a generous amount of cores and it has a capacity for superior performance in generally valuable applications. Graphical Processing Units (GPUs) have turned out to be imperative in giving handling power to superior performance applications. CUDA is a programming interface for GPU processing and it is an exclusive programming interface and collection of language extensions which works just on NVIDIA's GPUs. In this study, some of the image processing methods namely, Sobel, Prewitt and Robert's Cross edge detection are introduced and executed using different thread distributions and compared with the sequential implementation, i.e., single core CPU and multiple-core CPU. Execution outcomes show that critical speedup is accomplished with the usage of GPU as compared to single-core CPU and multiple-core CPU. It is also observed that the speedup increases with the increase in image size.