This paper presents an efficient image exploration scheme for the unshaped object using semantic modelling. The local regions of an image have been classified with respect to the frequency of occurrences. The semantic concept is evaluated using RGB histogram dissimilarity factor, overall dissimilarity factor and regional dissimilarity factor. The dissimilarities determine the local concept with accuracy up to 89.86% which is much higher than the existing techniques. The proposed algorithm also allows to ranks the unshaped objects according to their semantic similarity.