The comparison of ear recognition methods under different illumination effects and geometrical changes

Abstract

This paper presents the study of the permanence of the ear shape. The focus is on comparing ear recognition methods using images affected by illumination and geometrical changes. The main aim of the study is to determine the permanence of the ear shape and when does the ear stop developing. Whereas, the current stage aims to determine the most suitable method that can be used for ear recognition of young children that still under-go different geometrical changes and skin complexion changes. The suitable algorithm should be less sensitive to illumination and more sensitive to growth in order to be able to track significant changes of the ear caused by growth. Methods that are evaluated are the Histogram of Oriented Gradients (HOG), Patterns of Oriented Edge Map (POEM), Local Binary Patterns (LBP) and Gabor Filters. These methods were selected theoretically from the literature review as they were reported to show sensitivity to illumination and to geometrical changes. To perform the evaluation, 1000 ear images were generated from 100 ear images, 10 per each subject. For each subject, all 10 images have different illumination and another 10 have different geometrical changes. The results obtained show that a combination of HOG and LBP is suitable for ear recognition under geometrical and illumination changes.