一、<Bag-Of-Visual-Words and Spatial Extensions for Land-Use classification>
Future work point 1:
the color and BOVW approaches are othogonal in that the interest point descriptors are extracted using only the luminance channel so that a combined approach is likely to perform even better. This is possible future work.
Future work point2:
for sck+bovw, it is possibly using different sized dictionaries for each. For that, the SCK representation grows quadratically with the number of visual words but the plots indicate that a slightly larger dictionary should increase performance.
