Human visual perception is capable of much more challenging tasks than just object classification and detection. For example human can spot an abnormal/atypical image, and reason about what makes it strange. This task has not received enough attention in the artificial intelligence and computer vision community. There are various reasons that can make an image flagged as abnormal. In this talk we investigate the various reasons of abnormality in images in a more comprehensive way than have been done previously. We designed a human subject experiment to discover a coarse taxonomy of the reasons of abnormality. Our experiment showed that there are three major categories of abnormality: object-centric, scene-centric, and contextual. Based on this taxonomy, we propose
 computational models that can predict these different types of abnormality in images. We can reason about abnormality of images in terms of visual attributes. Visual attributes are semantically meaningful visual concept that have been used in object recognition and image description. We acquire this concept in our model to describe the exact reasons of abnormality in images and objects.