期刊名称：Journal of Theoretical and Applied Information Technology
印刷版ISSN：1992-8645
电子版ISSN：1817-3195
出版年度：2019
卷号：97
期号：3
页码：825-834
出版社：Journal of Theoretical and Applied
摘要：Human ears have shown significant robustness and distinctiveness in human biometric applications. Thus, many techniques have been proposed to recognize humans, based on the shape of their ears. In this research, Speeded-Up Robust Feature (SURF) technique is employed in an ear recognition method, which measures the similarity between the input ear image and the ear images of known individuals in a database. The performance of this method is optimized using Particle Swarm Optimizer (PSO), which is used to reduce the size of the descriptors generated by the SURF method, before used in similarity measurements. Any descriptor value that has negative or no influence on the similarity measurements is removed. The approach (SURF-PSO) has been able to improve the performance of the SURF method, by increasing the recognition accuracy and reducing the time required of measuring the similarities. The evaluation results using three datasets, show that the recognition accuracy of the SURF-based method is improved by 2.41%, while the average time per each similarity measurement has been reduced by 43.72%. The performance of the optimized method is compared to the performance of ear recognition using Feed-Forward Neural Network (FFNN) and Convolutional Neural Network (CNN) techniques, which are designed and implemented. Although the use of CNN has shown better performance, the classification approach is difficult to implement in real-life applications, according to the need to retrain the classifier upon any modification to the database of the known individuals, and also the topology of the classifying neural network is relevant to the number of individuals that it can recognize. However, the proposed method has shown better performance than the other techniques in the literature that are used for the same applications. where the performance optimization has reduced the gap between the SURF-PSO method and the CNN classification method , where the gap between CNN method and SURF method is 6% and decreased to 3.59% between CNN method and SURF-PSO .
