Year of Publication: Jun - 2012
Page Numbers: 91-101
Authors: Lailatul Qadri binti Zakariaa, Paul Lewisb, Wendy Hallc
Conference Name: The International Conference on Informatics and Applications (ICIA2012)
- Malaysia


Image analysis and in particular automatic image classification or annotation, often begins with feature extraction for content representation. The content representation can be based on low level and high level feature extraction. In this paper, we present a simple image classifier which attempts to classify building images based on their low level features which are colour and edges. The classifier is developed by using Bayesian Inference method provided by Infer.net tool. Equivalently, the classifier can be regarded as an annotator which aims to annotate images with a specific tag when appropriate. In this example, the image classifier is developed to identify and distinguish building and non images. The method could be extended into other classes such as mountains, beaches and forest. This image classification is important because it can enable users to retrieve images that may not be well tagged and also to annotate images with information that they may want to use for retrieval purposes but not necessarily for explicit annotation. The performances are assessed using confusion matrices and ROC curves.