Clothing analysis for subject identification and retrieval

Abstract

Soft biometrics offer several advantages over traditional biometrics.With given poor quality data, as in surveillance footage, most traditional biometrics lose utility, whilst the majority of soft biometrics is still applicable. Amongst many of a person’s descriptive features, clothing stands out as a predominant characteristic of their appearance. Clothing attributes can be effortlessly observable and described conventionally by accepted labels. Although there are many research studies on clothing attribute analysis, only few are concerned with analysing clothing attributes for biometric purposes. Hence, the use of clothing as a biometric for person identity deserves more research interest than it has yet received. This chapter provides extended analyses of soft clothing attributes and studies the clothing feature space via detailed analysis and empirical investigation of the capabilities of soft biometrics using clothing attributes in human identification and retrieval, leading to a perceptive guide for feature subset selection and enhanced performance. It also offers a methodology framework for soft clothing biometrics derivation and their performance evaluation.