spatial histograms are used to preserve some local information in the image.
instead of doing one histogram for the whole picture, you slice the image into n x n
smaller patches, like 7x7, and make a histogram for that pach only.
if you append those small histograms to a single one, that's your spatial histogram.
though it's all 'in one line', the local information (e.g, if something happened in the top left corner) will get saved this way.
you can do that with colors, binary patterns, or whatever featureset you can count into histograms
the lbp face recognition in opencv might me a nice example,
here's the paper. :
http://www.ee.oulu.fi/mvg/files/pdf/pdf_730.pdf
did opencv have any function to compute spatial histogram of video frames?