ABSTRACT

Annotating images automatically is needed for indexing and searching images in a big database. Image annotation can be considered as a multi-class classification problem, where an image may require one or more labels. In this work, One-Against-All (OAA) multi-class ANNs architecture approach is proposed to classifying into multi-class. The extracted features from color and edge descriptors of an image are used as input of the model. Some experiments were performed to achieve the optimal number of hidden neurons of each neural network that can classify its corresponding target successfully. The empirical results outperform

multiple label learning approach.

Keywords: Semantic scene classification, Multi-label classification, neural networks.