Clothing Color Based De-Identification

Abstract

Color of clothing can be easily used in the process of people identification as a non-biometric trait. This trait would be modified appropriately to preserve personal anonymity in social media or in video surveillance systems. This paper deals with a proposal of new method for clothing color de-identification in image records. The method consists of silhouette segmentation and clothing color segmentation algorithms. The de-identification is done by modifying components of HSV color space upon segmented clothing. The process of de-identification is reversible, i.e. original clothing color can be restored in case of transformation parameters that were used are known.