A comparison of object-based and contextual pixel-based classifications using high and medium spatial resolution images

Object-based classification has demonstrated numerous advantages over non-contextual pixel-based classification due to its capability of modelling spatial information through image segmentation. Similarly, contextual pixel-based classification can also incorporate spatial information among neighbouring pixels to improve the performance of non-contextual pixel-based classification. However, to our knowledge, no study has compared object-based approaches with contextual pixel-based approaches for image classification. In this letter, we compared an object-based approach using a segmentation algorithm embedded in eCognition with a contextual pixel-based approach using Markov random fields. The performances were evaluated with a high spatial resolution image (3 m) and a medium spatial resolution image (30 m) using various thematic and geometric accuracy indices. The results showed that the classification accuracy of the contextual pixel-based approach is higher than the object-based approach on both images, and the values of geometric indices for the two approaches are comparable.