A configurable architecture for fast moments computation

Kah Hyong Chang, Raveendran Paramesran

Research output: Contribution to journalArticleResearchpeer-review

1 Citation (Scopus)

Abstract

In this paper, we present a single-chip architecture for generating a full set of geometric moments using digital filters. Other types of moments such as Zernike and Tchebichef moments can also be implemented. The architecture can be configured for any order of geometric moments and image spatial resolution at run time. The use of a single-scaler method and reusable hardware resources enables higher order moments to be computed. The incorporation of two-level pipelining and masking techniques further increases the throughput. Realized in a field-programmable gate array, the design is capable of processing sixty 512 × 512 8-bit-pixel images per second at 20 MHz, generating (59 + 59) orders of geometric moments (3,600 moments). The maximum round-off error is approximately 1 %.

Original languageEnglish
Pages (from-to)179-186
Number of pages8
JournalJournal of Signal Processing Systems
Volume78
Issue number2
DOIs
Publication statusPublished - Feb 2013
Externally publishedYes

Keywords

  • Configurable
  • Digital filters
  • Field-Programmable Gate Array (FPGA)
  • High-order
  • Image processing
  • Moments
  • Real-time

Cite this