High-throughput image rotation using sign-prediction based redundant CORDIC algorithm

Real-time image rotation forms a core operation in many applications such as medical image processing and computer vision. High-throughput computations for image rotation are a common requirement in real-time image processing. A VLSI design for image rotation that employs CORDIC was discussed in Ghosh et al. (1994). In this paper, we propose novel techniques to increase the throughput of the CORDIC computations, thereby notably improving the overall performance of the rotation unit with acceptable increase in VLSI area. An efficient sign-prediction method called BTSP (Binary-Tree based Sign Prediction) is proposed in our redundant CORDIC system to eliminate the sign detection process. Our investigations show that the BTSP based CORDIC algorithm for realtime rotation of a 512/spl times/512-pixel image, has a significant speed-up over existing redundant CORDIC methods with reduced hardware area.