Ear Detection and Extraction Using a Developed Features Descriptor and Geometric Transformation

Ear recognition for biometric identification purposes is increasingly popular in several recent fields. However, a lack of searches focuses on ear area detection and extraction from the whole scene in real-time conditions. This paper introduces SURFR as a proposed method to gather some prominent features of the human pinna outer helixes, which have been stored as a feature vectors dataset, to detect, extract, and then pick up the desired area of the ear to be identified later. The interest points of features were extracted using the SURF descriptor for both the stored outer-helix dataset and the captured side-face image. Later, the matched features of the stored dataset and the captured person's side-face image have been gathered in two matrices, which were entered to calculate the geometric transformation based on the RANSAC method. The proposal has experimented with UBEAR, AMI, UMSIT, and USTB datasets, in addition to our image data. By leveraging our proposal, results have achieved the real-time demands with an accuracy rate of 95–97 % for detection and extraction respectively.