Use of Nature-inspired Meta-heuristics for Handwritten Digits Recognition

Character recognition is an important task in pattern analysis that aims to give significance to handwritten data without users’ intervention. Although, an intensive research has been devoted to this problem, it remains a challenging task as humans need to interact with computer in the easiest way. This work attempts to incorporate some meta-heuristics as guidelines searching for the best solution of handwritten digits recognition problem namely the particle swarm optimizer and variations of the bees’ algorithm. The bees’ algorithm is a variant of evolutionary optimization that take inspiration from the foraging behavior of honey bees where individuals called bees are used to perform a neighborhood search in joint with a random search as an attempt to achieve a good balance between exploration and exploitation abilities. We show that this method can be adapted to handwritten characters recognition and can be effectively combined with a neural network classifier which results in a good quality on a wide range of real data against that of the k-nearest neighbour classifier and the back-propagation training algorithm.