Pipelined IC architecture for radon transform computations in a multiprocessor array.

摘要

The amount of data generated by CT scanners is enormous, making the reconstruction operation slow, especially for 3-D and limited-data scans requiring iterative algorithms. The Radon transform and its inverse, commonly used for CT image reconstruction from projections, are computationally burdensome for today's single-processor computer architectures. If the processing times for the forward and inverse Radon transforms were comparatively small, a large set of new CT algorithms would become feasible, especially those for 3-D and iterative tomographic image reconstructions. In addition to image reconstruction, a fast ''Radon Transform Computer'' could be naturally applied in other areas of multidimensional signal processing including 2-D power spectrum estimation, modeling of human perception, Hough transforms, image representation, synthetic aperture radar processing, and others. A high speed processor for this operation is likely to motivate new algorithms for general multidimensional signal processing using the Radon transform. In the proposed workshop paper, we will first describe interpolation schemes useful in computation of the discrete Radon transform and backprojection and compare their errors and hardware complexities. We then will evaluate through statistical means the fixed-point number system required to accept and generate 12-bit input and output data with acceptable error using the linear interpolation scheme selected. These results set some of the requirements that must be met by our new VLSI chip architecture. Finally we will present a new unified architecture for a single-chip processor for computing both the forward Radon transform and backprojection at high data rates. 3 refs., 2 figs.