Fast and efficient FPGA implementation of connected operators

The Connected Component Tree (CCT)-based operators play a central role in the development of new algorithms related to image processing applications such as pattern recognition, video-surveillance or motion extraction. The CCT construction, being a time consuming task (about 80% of the application time), these applications remain far-off mobile embedded systems. This paper presents its efficient FPGA implementation suited for embedded systems. Three main contributions are discussed: an efficient data structure proposal adapted to representing the CCT in embedded systems, a memory organization suitable for FPGA implementation by using on-chip memory and a customizable hardware accelerator architecture for CCT-based applications.

► Connected Component Tree (CCT) data structure for efficient data mapping in memory.
► On-chip memory organization suited for CCT processing.
► Hardware accelerator architecture for CCT based applications.