An Improved Radial Basis Function Neural Network

Only available on StudyMode
  • Download(s) : 82
  • Published : April 5, 2013
Open Document
Text Preview
An Improved Radial Basis Function Neural Network
based on a Cooperative Coevolutionary Approach for
Handwritten Digits Recognition
Salima Nebti

Abdellah Boukerram

Department of computer science
Ferhat Abbas University
Sétif 19000, Algeria
snebti@live.fr

Department of computer science
Ferhat Abbas University
Sétif 19000, Algeria
boukerram@hotmail.com

Abstract—Co-evolutionary algorithms are a class of adaptive search meta-heuristics inspired from the mechanism of reciprocal benefits between species in nature. The present work proposes a cooperative co-evolutionary algorithm to improve the

performance of a radial basis function neural network (RBFNN) when it is applied to recognition of handwritten Arabic digits. This work is in fact a combination of ten RBFNNs where each of them is considered as an expert classifier in distinguishing one digit from the others; each RBFNN classifier adapts its input features and its structure including the number of centres and their positions based on a symbiotic approach. The set of

characteristic features and RBF centres have been considered as dissimilar species where each of them can benefit from the other, imitating in a simplified way the symbiotic interaction of species in nature. Co-evolution is founded on saving the best weights and centres that give the maximum improvement on the sum of

squared error of each RBFNN after a number of learning
iterations. The results quality has been estimated and compared to other experiments. Results on extracted handwritten digits from the MNIST database show that the co-evolutionary
approach is the best.
Keywords-Handwritten digits recognition, particle swarm
optimization, Co-evolution, radial basis neural networks.

I.

INTRODUCTION

Character recognition is an active and important area of
research in the field of data processing since it allows an easier interaction with computer machines. An optical character
recognition system (OCR) treats digital images coming from
scanner or camera, and then produces a textual file of printed characters. Generally, an OCR follows three necessary steps: The first is "preprocessing": which includes character
adjustments and text segmentation into lines and lines into
words or isolated characters. Segmentation constitutes a hard task in the preprocessing step since it is difficult to determine where begins and where finishes a word without recognizing
its characters or the word itself due to characters overlapping. Once the character is preprocessed, the next step extracts its characteristic features. The extracted features are then used for classification or recognition. In this step, a character is

compared with many known forms, and then the nearest form

is retained.
After classification, other post-processing
functions can be achieved to reduce the recognition errors.
Post-processing generally uses linguistic and contextual rules such as dictionaries of words, syllables, or trigrams to
eliminate the incorrect solutions [12].
An OCR can be holistic or analytical. In the holistic case,
the word is regarded as a unit to be recognized, horizontal and vertical profiles are often used as characteristic features. This approach is effective because it prevents segmentation errors as it does not require characters segmentation. However it

needs a big size of lexicon and training data, as a result it needs a very long computing time for training. The analytical approach considers each character or each characteristic
feature as a unit to be recognized. Therefore, it is necessary to segment words into characters as a result the recognition rate decreases due to segmentation errors which are generally
reduced by using hidden markov models (HMMs) [13].
Character recognition methods can be distinguished in two
main classes: structural and statistical methods. In the first class, each character can be described by structural features. The recognition process uses...
tracking img