Human Centric Recognition of 3D Ear Models

Abstract

Comparing ear photographs is considered to be an important aspect of disaster victim identification and other forensic and security applications. An interesting approach concerns the construction of 3D ear models by fitting the parameters of a ‘standard’ ear shape, in order to transform it into an optimal approximation of a 3D ear image. A feature list is then extracted from each 3D ear model and used in the recognition process. In this paper, we study how the quality and usability of a recognition process can be improved by computational intelligence techniques. More specifically, we study and illustrate how bipolar data modelling and aggregation techniques can be used for improving the representation and handling of data imperfections. A novel bipolar measure for computing the similarity between corresponding feature lists is proposed. This measure is based on the Minkowski distance, but explicitly deals with hesitation that is caused by bad image quality. Moreover, we investigate how forensic expert knowledge can be adequately reflected in the recognition process. For that reason, a hierarchically structured comparison technique for feature sets and other characteristics is proposed. Comparison results are expressed by bipolar satisfaction degrees and properly aggregated to an overall result. The benefits and added value of the novel technique are discussed and demonstrated by an illustrative example.

Keywords

• Ear photograph recognition
• computational intelligence
• bipolarity
• aggregation