OPTIMIZATION OF FPGA-BASED PROCESSOR ARCHITECTURE FOR SOBEL EDGE DETECTION OPERATOR

This dissertation introduces an optimized processor architecture for Sobel edge detection operator on field programmable gate arrays (FPGAs). The processor is optimized by the use of several optimization techniques that aim to increase the processor throughput and reduce the processor logic utilization and memory usage. FPGAs offer high levels of parallelism which is exploited by the processor to implement the parallel process of edge detection in order to increase the processor throughput and reduce the logic utilization. To achieve this, the proposed processor consists of several Sobel instances that are able to produce multiple output pixels in parallel. This parallelism enables data reuse within the processor block. Moreover, the processor gains performance with a factor equal to the number of instances contained in the processor block. The processor that consists of one row of Sobel instances exploits data reuse within one image line in the calculations of the horizontal gradient. Data reuse within one and multiple image lines is enabled by using a processor with multiple rows of Sobel instances which allow the reuse of both the horizontal and vertical gradients. By the application of the optimization techniques, the proposed Sobel processor is able to meet real-time performance constraints due to its high throughput even with a considerably low clock frequency. In addition, logic utilization of the processor is low compared to other Sobel processors when implemented on ALTERA Cyclone II DE2-70.