Department of Computer Science and Technology

Technical reports

Pipelined image processing for pattern recognition

A. Daniel Hall

July 2016, 121 pages

This technical report is based on a dissertation submitted October 1991 by the author for the degree of Doctor of Philosophy to the University of Cambridge, Queen’s College.

DOI: 10.48456/tr-892


Image processing for pattern recognition is both computationally intensive and algorithmically complex. The objective of the research presented here was to produce a fast inexpensive image processor for pattern recognition. This objective has been achieved by separating the computationally intensive pixel processing tasks from the algorithmically complex feature processing tasks.

The context for this work is explored in terms of image processor architecture, intermediate-level image processing tasks and pattern recognition.

A new language to describe pipelined neighbourhood operations on binary images (‘PiNOLa’: Pipelined Neighbourhood Operator Language) is presented. PiNOLa was implemented in Modula-2 to provide an interactive simulation system (‘PiNOSim’: Pipelined Neighbourhood Operator Simulator). Experiments using PiNOSim were conducted and a design for a topological feature extractor was produced.

A novel algorithm for connected component labelling in hardware is presented. This algorithm was included in the PiNOSim program to enable the component labelling of features extracted using the language. The component labelling algorithm can be used with the topological feature extractor mentioned above. The result is a method of converting a binary raster scan into a stream of topological features grouped by connected object.

To test the potential performance of a system based on these ideas, some hardware (‘GRIPPR’: Generic Real-time Image Processor for Pattern Recognition) was designed. This machine was implemented using standard components linked to a PC based transputer board. To demonstrate an application of GRIPPR an Optical Character Recognition (OCR) system is presented. Finally, results demonstrating a continuous throughput of 1500 characters/second are given.

Full text

PDF (8.7 MB)

BibTeX record

  author =	 {Hall, A. Daniel},
  title = 	 {{Pipelined image processing for pattern recognition}},
  year = 	 2016,
  month = 	 jul,
  url = 	 {},
  institution =  {University of Cambridge, Computer Laboratory},
  doi = 	 {10.48456/tr-892},
  number = 	 {UCAM-CL-TR-892}