Edge Detection Using Cellular Automata - Cellular Automata in Image Processing and Geometry - page 82

Chapter 5
Edge Detection Using Cellular Automata

Paul L. Rosin and Xianfang Sun

Abstract. Edge detection has been a long standing topic in image processing, gen-
erating hundreds of papers and algorithms over the last 50 years. Likewise, the topic
has had a fascination for researchers in cellular automata, who have also developed
a variety of solutions, particularly over the last ten years. CA based edge detection
has potential benefits over traditional approaches since it is computationally effi-
cient, and can be tuned for specific applications by appropriate selection or learning
of rules. This chapter will provide an overview of CA based edge detection tech-
niques, and assess their relative merits and weaknesses. Several CA based edge de-
tection methods are implemented and tested to enable an initial comparison between
competing approaches.

5.1 Introduction

Edge detection is a fundamental tool for computer vision systems. The original use
of boundaries detected in a scene was for object detection, as they provide a set of
features suitable for model matching. However, edge maps are nowadays applied
to a host of different ways, such as preventing removal of significant structures in
anisotropic blurring [23], indicating salient regions [33], selecting redundant seams
in image resizing [1], image registration [25], depth from focus [8], extended depth
of field, etc.
From Roberts' early work [31] in 1963 through to Canny's influential pa-
per [4] in 1986, most methods have used first-order derivatives to estimate edge
magnitude and orientation. Subsequently, the edgels (i.e. edge pixels) are often
thresholded, linked, and thinned. Edge detection is normally carried out using lo-
cal operators, and since edges may not be locally distinct in the image this means
that edge maps are prone to fragmentation. Thus, research into new methods of edge

Paul L. Rosin · Xianfang Sun
School of Computer Science & Informatics, Cardiff University, Cardiff, CF24 3AA, UK
e-mail: {Paul.Rosin,Xianfang.Sun}@cs.cf.ac.uk