"An FPGA Co-Processor for Statistical Pattern Recognition Applications"

Jason C. Isaacs and Simon Y. Foo
Florida A&M University - Florida State University

Abstract

Pattern recognition applications are pervasive in today’s data driven technology, from simple optical character recognition problems to complex autonomous defense applications. The need for near real-time computation stresses the importance of fast algorithms and parallel/pipelined implementations. Since it is highly impractical to cart BlueGene along, the necessity for a small portable device increases, especially for stringent weight constraint applications. We have implemented L spatial filters derived from the Leung-Malik filter bank and S spatial filters derived from the Schmid filter bank for an object recognition application. Typically, L ranges from two to sixty-two depending on the number of orientations and the resolution scale. For the Schmid filters, S ranges from one to thirteen. All filters have the same support, i.e., they are N x N. Therefore, the filter coefficients are stored in look-up tables and called when necessary. The number of lookup tables varies with L and S; in addition, the size of the tables varies with support. These filters are typically implemented in floating-point; however, with limited floating-point resources they are implemented here in a fixed-point representation. The target device for this application is a Virtex-II PCI development board. Significant speed-up demonstrates the utility of stand-alone FPGA devices as well as FPGA co-processors for computationally intensive applications.

Keywords: Pattern recognition, spatial filters, Schmid Filters, FPGA co-processor

References:

  1. T. Leung and J. Malik. Representing and recognizing the visual appearance of materials using three-dimensional textons. International Journal of Computer Vision, 43(1):29-44, June 2001.
  2. C. Schmid. Constructing models for content-based image retrieval. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, volume 2, pages 39-45,

2005 MAPLD International Conference Home Page