SVM BASED INDOOR/MIXED/OUTDOOR CLASSIFICATION FOR DIGITAL PHOTO ANNOTATION IN A UBIQUITOUS COMPUTING ENVIRONMENT

Abstract

This paper extends our previous framework for digital photo annotation by adding noble approach of indoor/mixed/outdoor image classification. We propose the best feature vectors for a support vector machine based indoor/mixed/outdoor image classification. While previous research classifies photographs into indoor and outdoor, this study extends into three types, including indoor, mixed, and outdoor classes. This three-class method improves the performance of outdoor classification. This classification scheme showed 5-10% higher performance than previous research. This method is one of the components for digital image annotation. A digital camera or an annotation server connected to a ubiquitous computing network can automatically annotate captured photos using the proposed method.