In this paper, a computationally efﬁcient human recognition technique has been proposed using Unique Mapped Real Transform (UMRT) from ear biometric modality. This technique is time saving as well as robust against illumination changes and rotations. First, the input ear image is preprocessed to improve its overall visual appearance. The desired ear region is segmented out from the preprocessed image using constrained Delaunay triangulation segmentation technique. A computationally efﬁcient and robust UMRT is then used to extract feature vectors which uniquely represent ear images of different persons. The performance of proposed feature vector extraction is studied by testing the feature vectors using the KNN classiﬁer and Euclidean distance classiﬁer. The proposed ear recognition technique is also compared with Uniform Local Binary Pattern (ULBP) based technique. Testing is carried out using IIT Delhi and internal GEAR ear database images and the results are encouraging.
Digital Object Identifier (DOI)
Sowmyalakshmi, R. and M. Girirajkumar, S.
"A New Technique for Computationally Efﬁcient Human Recognition,"
Applied Mathematics & Information Sciences: Vol. 11:
2, Article 14.
Available at: https://digitalcommons.aaru.edu.jo/amis/vol11/iss2/14