High level semantic image recognition and classification is a challenging task and currently is a very active research domain. Computers struggle with the high level task of identifying objects and scenes within digital images accurately in unconstrained environments. In this paper, we present experiments that aim to overcome the limitations of computer vision algorithms by combining them with novel contextual based features to describe geo-tagged imagery. We adopt a machine learning based algorithm with the aim of classifying classes of geographical landmarks within digital images. We use community contributed image sets downloaded from Flickr and provide a thorough investigation, the results of which are presented in an evaluation section.