Exploring neighborhood and spatial information for improving scene classification

• Distant neighbors to the non-parametric transform improve scene representation.
• We consider distant neighbors without explode the size of the feature vector.
• The space arrangement of statistical information enhances scene representation.
• We consider contrast information of the pixels in a 3×33×3 window.

A good image descriptor is essential for the scene classification task. This work proposes an improvement for the Contextual Mean Census Transform (CMCT), an image descriptor, obtained by adding information from distant neighbors to the non-parametric transform computation and spatial information. We also combine the new descriptor with the gist and Spatial Modified Census Transform (Spatial MCT) descriptors to improve classification performance. Experimental results on four commonly used datasets demonstrate that the proposed methods can achieve performance competitive with previous methods.