Feature selection using dynamic binary particle swarm optimization for Arabian horse identification system

Abstract

Being able to formally identify horses is crucial for many reasons like biosecurity and regulatory risks, fairness in competition to ensure the proper horse and owner are competing in an event, retrieval after theft and medical record management. Among different horses types Arabian horse is one of the top ten popular horse breeds all over the world. It is the most expensive horse in the market. Such horse identification system can be based on biometric parameters. This work aims to introduce a novel method of periocular region segmentation using Otsu based method in combination with Improved Fruit Fly Optimization Algorithm (IFOA) followed by a feature extraction and selection phase for Arabian horse identification system. The segmented horse periocular region is subjected to texture analysis using Gabor filter and discrete cosine transform for proper feature extraction. A proper Feature Selection step is performed with the aim of selecting optimum features. Such optimal set of features will be used later in Arabian horse identification and recognition system. Such optimal feature selection is achieved using Dynamic Binary Particle Swarm Optimization.