This work establishes the high value of ear images for personal identification from mugshot data, using the NIST database of police mugshots. It starts with a method for boundary analysis based on two innovations. First edge analysis is performed only along rays emanating from a point near the center of the ear. This is much faster than applying a Canny edge detector to the entire image. The second innovation is the use of interpretation breeding. Two distinct methods are used to find the ear boundary, and these interpretations are merged in order to find the best boundary. This results in good segmentation for well over 70 % of the images. The segmented ears are cut out from the original profile, and standardized in several ways to compensate for image variations. For identification, a neural network is used to compute a composite distance criterion. Individual distances include one based on components of an eigenear basis similar to Pentland's eigenfaces, and one based on comparison of the most robust portion of the boundary curve. The best match to a random query is found 58 % of the time, and the correct match is among the top five 77 % of the time. These results compare favorably with those for frontal images from the NIST mugshot database.
face recognition, image processing, neural networks
Personal Identification From Mugshot Ear Images, - 6130, National Institute of Standards and Technology, Gaithersburg, MD, [online], https://doi.org/10.6028/NIST.IR.6130
(Accessed June 3, 2023)