Attire-based photo annotation

Many commercial and free photo organizers allow users to manage their personal photo collections by grouping them into folder, album and by dates. However, these are done manually by dragging and dropping these photos into named albums and star-marked folders. In this paper, we proposed automated photo organization by labeling who are in the images as well as group them into events accordingly. This process is known as automated photo annotation. In most automated photo annotation, basic metadata such as time and date are typically used to group them by events. We took one step further by incorporating the clothing colour feature to perform automated photo annotation. A total of 175 photographs comprising 138 frontal faces are selected from events such as Eid celebration, birthday celebration and outdoor occasions. The face features are described using eigenfaces, while the clothing colours are represented using La*b* colour histogram. Julian date format is added into the feature vector and three metric-distance similarity measures are compared for performance. Our experiment showed that the highest person annotation rate was achieved by combination of face and clothing feature at 82.5 percent. On the other hand, event is best annotated using combination of face and time features.