In the field of ear biometrics, despite recent developments, there are no freely available tools or databases captured in the wild that would ease the comparison of methods for ear biometric recognition. A new ear database captured in the wild was developed as a part of this thesis as well as a new toolbox for ear recognition. Rotation invariant local phase quantization method was also applied, for the first time in the field of ear biometrics, together with a fusion of this method with vectors of normalized images. Results are comparable but not better than the state-of-the-art, however, with the fusion we have achieved better results than the rotation invariant local phase quantization alone. The developed toolbox for ear recognition presents a new step towards standardization in the field of ear recognition and tests on the ear database have shown that the database presents greater challenge than the existing ones, but it is still comparable, which makes it suitable for further use.
|
