Unsupervised classification of events: A semantic rule based on color moments of background and foreground method

Abstract

Identification/labeling of the events is still an open problem for researchers due to the problem of semantic gap between the low level features and human perception though it is essential in solving the fundamental problem of automatic albuming, managing digital photography and performance of their retrieval system. Hence, in this paper, we present novel unsupervised semantic rule (USR) based method to label the events such as indoor (wedding) and outdoor (picnic) by establishing semantic relationship between successive background and foreground information of digital photographs. In order to derive a semantic rule, we have used color moments as features and Euclidean distance measure as the nearest neighbor. The rule is defined as the mean values of background information of images, which is lesser than mean value of foreground information in case of indoor event. Similarly, the mean value of background information is higher than the mean values of foreground information in case of outdoor images, This is true because of the fact that background information of Indoor photographs is constant over a long time in a whole event, whereas foreground information varies during the same time. Similarly, the background information varies in case of outdoor photographs but foreground information is almost constant. Experimental results of the proposed method are compared with results of the existing methods to bring the superiority of the proposed method in terms of robustness, simplicity, and accuracy in labeling the events without a priori knowledge.