Skip to main content
U.S. Department of Energy
Office of Scientific and Technical Information

Parallel algorithms for one-dimensional infinite impulse response filters and two-dimensional image matching

Thesis/Dissertation ·
OSTI ID:5585744
Parallel algorithms are very important to such areas as signal processing, image processing, and computer vision. For signal processing, sampling times can be as small as ten nanoseconds. For image processing and computer vision, it may take minutes of cpu time to process one single image. The combination of VLSI and parallel algorithms provides a good way to solve the computational bottlenecks that exist in these areas. First, for signal processing, the author has studied the problem of computing the equations for one-dimensional Infinite Impulse Response (IIR) filters and have developed a new parallel algorithm. Second, for image processing and computer vision, the author has studied the problem of two-dimensional image recognition and have developed two warping-based image recognition algorithms. Dynamic programming is the method used to accumulate the local results into a global result. Because these algorithms are computationally intensive, special VLSI architectures have been designed. The dynamic space-warping algorithm is the first warping algorithm that has been developed. It finds the minimum distance between two two-dimensional areas based on compression and expansion. The dynamic line-warping algorithm is the second warping algorithm that has been developed. The major function of this algorithm is to find the minimum distance between a one-dimensional area and a two-dimensional area based on compression and expansion. It can be implemented directly by a three-dimensional processor array. For VLSI implementation, a two-dimensional architecture has been derived from the three-dimensional architecture. By running simulators on the Intel hypercube machine iPSC2, experimental results have shown that these algorithms can recognize both character and picture images.
Research Organization:
Pennsylvania State Univ., Middletown, PA (United States)
OSTI ID:
5585744
Country of Publication:
United States
Language:
English