EvoIASP2003
5th European Workshop on Evolutionary Computation in Image Analysis and Signal Processing

The 12 papers accepted for last year's EvoIASP workshop are available in Springer's LNCS series, volume 2279.

Previous editions:
Göteborg, Sweden, 1999
Edinburgh, UK, 2000
Lake Como, Italy, 2001
Kinsale, Ireland, 2002

Introduction

EvoIASP2003 is the fifth workshop of the EvoNet working group on image analysis and signal processing.

Evolutionary algorithms have been shown to be an effective tool for the development of image analysis and signal processing systems (both software and hardware) in complex domains of high industrial and social relevance.

EvoIASP 2003 aims to foster closer interaction between members of the evolutionary computing, image analysis and signal processing communities. The workshop provides an opportunity for European and non-European researchers, as well as people from industry, to present their research and to discuss the latest developments and applications.

The workshop proceedings will be published by Springer in the LNCS series and will be available at the workshop.

Topics of interest

Topics of interest include, but are not limited to:

• applications of evolutionary computation to real-life IASP problems
• evolvable vision and signal processing hardware
• evolutionary pattern recognition
• hybrid architectures for machine vision and signal processing including evolutionary components
• theoretical developments
• comparisons between different evolutionary techniques and between evolutionary and non-evolutionary techniques in IASP applications
• time series analysis by means of EC techniques.

Draft: subject to change

See also: Programme overview

The EvoWorkshops2003 proceedings will be
published by Spinger as part of their
Lecture Notes in Computer Science series.

Accurate L-corner Measurement using USEF Functions and Evolutionary Algorithms
Olague G, n Hernandez B, Dunn E
Corner feature extraction is studied in this paper as a global optimization problem. We propose a new parametric corner modeling based on a Unit Step Edge Function (USEF) that defines a straight line edge. This USEF function is a distribution function, which models the optical and physical characteristics present in digital photogrammetric systems. We search model parameters characterizing completely single gray-value structures by means of least squares fit of the model to the observed image intensities. As the identification results relies on the initial parameter values and as usual with non-linear cost functions in general we cannot guarantee to find the global minimum. Hence, we introduce an evolutionary algorithm using an affine transformation in order to estimate the model parameters. This transformation encapsulates within a single algebraic form the two main operations, mutation and crossover, of an evolutionary algorithm. Experimental results show the superiority of our L-corner model applying several levels of noise with respect to simplex and simulated annealing.
EvoIASP Session 1: Computer Vision: April 14, 1130-1300

Anticipating Bankruptcy Reorganisation from Raw Financial Data using Grammatical Evolution
Brabazon A, O'Neill M
This study using Grammatical Evolution, constructs a series of models for the prediction of bankruptcy, employing information drawn from financial statements. Unlike prior studies in this domain, the raw financial information is not preprocessed into pre-determined financial ratios. Instead, the ratios to be incorporated into the predictive rule are evolved from the raw financial data. This allows the creation and subsequent evolution of alternative ratio-based representations of the financial data. A sample of 178 publically quoted, US firms, drawn from the period 1991 to 2000 are used to train and test the model. The best evolved model in each time period correctly classified 78 (70)% of the firms in the out-of-sample validation set, one (three) year(s) prior to failure. The utility of a number of different Grammars for the problem domain is also examined.
EvoIASP Session 3: Miscellanea: April 14, 1600-1800

Evolutionary approach to discovery of classification rules from remote sensing images
Korczak J, Quirin A
In this article a new method for classification of remote sensing images is described. For most applications, these images contain voluminous, complex, and sometimes noisy data. For the approach presented herein, image classification rules are discovered by an evolution-based process, rather than by applying an a priori chosen classification algorithm. During the evolution process, classification rules are created using raw remote sensing images, the expertise encoded in classified zones of images, and statistics about related thematic objects. The resultant set of evolved classification rules are simple to interpret, efficient, robust and noise resistant. This evolution-based approach is detailed and validated based on remote sensing images covering not only urban zones of Strasbourg, France, but also vegetation zones of the lagoon of Venice.
EvoIASP Session 3: Miscellanea: April 14, 1600-1800

GAME-HDL: Implementation of Evolutionary Algorithms using Hardware Description Languages
Drechsler R, Drechsler N
Evolutionary Algorithms (EAs) have been proposed as a very powerful heuristic optimization technique to solve complex problems. Many case studies have shown that they work very efficient on a large set of problems, but in general the high qualities can only be obtained by high run time costs. In the past several approaches based on parallel implementations have been studied to speed up EAs. In this paper we present a technique for the implementation of EAs in hardware based on a the concept of reusable modules. These modules are described in a Hardware Description Language (HDL). The resulting ``hardware EA'' can be directly synthesized and mapped to Application Specific Integrated Circuits (ASICs) or Field Programmable Gate Arrays (FPGAs). This approach finds direct application in signal processing, where hardware implementations are often needed to meet the run time requirements of a real­time system. In our prototype implementation we used VHDL and synthesized an EA for solving the OneMax problem. Simulation results show the feasibility of the approach. Due to the use of a standard HDL, the components can be reused in the form of a library.
EvoIASP Session 3: Miscellanea: April 14, 1600-1800

Hybrid Evolution Strategy-Downhill Simplex Algorithm for Inverse Light Scattering Problems
Macias D, Olague G, Mendez
The rough surface inverse scattering problem is approached with a combination of evolutionary strategies and the simplex method. The surface, assumed one-dimensional and perfectly conducting, is represented using spline curves. Starting from rigorously calculated far-field angle-resolved scattered intensity data, we search for the optimum profile using the evolutionary strategies (μ/ρ,+λ). After a fixed number of iterations, the best surface is finally recovered with the downhill simplex method. Aspects of the convergence and lack of uniqueness of the solution are discussed.
EvoIASP Session 3: Miscellanea: April 14, 1600-1800

Mobile Robot Sensor Fusion Using Flies
Boumaza A, Louchet J
The "Fly algorithm" is a fast artificial evolution-based image processing technique. Previous work has shown how to process stereo image sequences and use the evolving population of "flies" as a continuously updated representation of the scene for obstacle avoidance in a mobile robot. In this paper, we show that it is possible to use several sensors providing independent information sources on the surrounding scene and the robot's position, and fuse them through the introduction of corresponding additional terms into the fitness function. This sensor fusion technique keeps the main properties of the fly algorithm: asynchronous processing, no low-level image pre-processing or costly image segmentation, fast reaction to new events in the scene. Simulation test results are presented.
EvoIASP Session 1: Computer Vision: April 14, 1130-1300

Multiple Genetic Snakes for Bone Segmentation
Ballerini L, Bocchi L
Clinical assessment of skeletal age is a frequent, but yet difficult and time-consuming task. Automatic methods which estimate the skeletal age from a hand radiogram are currently being studied. This work presents a method to segment each bone complex in the radiogram, using a modified active contour approach. Each bone is modelled by an independent contour, while neighbouring contours are coupled by an elastic force. The optimization of the contour is done using a genetic algorithm. Experimental results, carried out on a portion of the whole radiogram, show that coupling of deformable contours with genetic optimization allows to obtain an accurate segmentation.
EvoIASP Session 1: Computer Vision: April 14, 1130-1300

On Two Approaches to Image Processing Algorithm Design for Binary Images using GP
Quintana M, Poli R, Claridge E
In this paper we describe and compare two different approaches to design image processing algorithms for binary images using Genetic Programming (GP). The first approach is based on the use of mathematical morphology primitives. The second is based on Sub-Machine-Code GP, a technique to speed up and extend GP based on the idea of exploiting the internal parallelism of sequential CPUs. In both cases the objective is to find programs which can transform binary images of a certain kind into other binary images containing just a particular characteristic of interest. In particular, here we focus on the extraction of three different features in music sheets.
EvoIASP Session 2: Genetic Programming: April 14, 1400-1530

Pixel Statistics and False Alarm Area in Genetic Programming for Object Detection
Zhang M, Andreae P, Pritchard M
This paper describes a domain independent approach to the use of genetic programming for object detection problems. Rather than using raw pixels or high level domain specific features, this approach uses domain independent statistical features as terminals in genetic programming. Besides position invariant statistics such as mean and standard deviation, this approach also uses position dependent pixel statistics such as moments and local region statistics as terminals. Based on an existing fitness function which uses linear combination of detection rate and false alarm rate, we introduce a new measure called "false alarm area" to the fitness function. In addition to the standard arithmetic operators, this approach also uses a conditional operator 'if' in the function set. This approach is tested on two object detection problems. The experiments suggest that position dependent pixel statistics computed from local (central) regions and nonlinear condition functions are effective to object detection problems. Fitness functions with false alarm area can reflect the smoothness of evolved genetic programs. This approach works well for detecting small regular multiple class objects on a relatively uncluttered background.
EvoIASP Session 2: Genetic Programming: April 14, 1400-1530

Restoration of Old Documents with Genetic Algorithms
Rivero D, Vidal R, Dorado J, Rabunal J, Pazos A
Image recognition is a problem present in many real­world applications. In this paper we present an application of genetic algorithms (GAs) to solve one of those problems: the recovery of a deteriorated old document from the damages caused by centuries. This problem is particularly hard because these documents are affected by many aggressive agents, mainly by the humidity caused by a wrong storage during many years. This makes this problem unaffordable by other image processing techniques, but results show how GAs can successfully solve this problem.
EvoIASP Session 3: Miscellanea: April 14, 1600-1800

The Effectiveness of Cost Based Subtree Caching Mechanisms in Typed Genetic Programming for Image Segmentation
Roberts M
Genetic programming (GP) has long been known as a computationally expensive optimisation technique. When evolving imaging operations, the processing time increases dramatically. This work describes a system using a caching mechanism which reduces the number of evaluations needed by up to 66 percent, counteracting the effects of increasing tree size. This results in a decrease in elapsed time of up to 52 percent. A cost threshold is introduced which can guarantee a speed increase. This caching technique allows GP to be feasibly applied to problems in computer vision and image processing. The trade-offs involved in caching are analysed, and the use of the technique on a previously time consuming medical segmentation problem is shown.
EvoIASP Session 2: Genetic Programming: April 14, 1400-1530

Chair

Stefano Cagnoni <cagnoni@ce.unipr.it>

Programme committee

• Giovanni Adorni (Italy)
• Lucia Ballerini (Sweden)
• Wolfgang Banzhaf (Germany)
• Dario Bianchi (Italy)
• Alberto Broggi (Italy)
• Stefano Cagnoni (Italy)
• Ela Claridge (UK)
• Marc Ebner (Germany)
• Terry Fogarty (UK)
• Daniel Howard (UK)
• Mario Koeppen (Germany)
• Evelyne Lutton (France)
• Peter Nordin (Sweden)
• Gustavo Olague (Mexico)
• Riccardo Poli (UK)
• Conor Ryan (Ireland)
• Jim Smith (UK)
• Giovanni Squillero (Italy)
• Andy Tyrrell (UK)
• Hans-Michael Voigt (Germany)