Background texture classification of portrait images
I am working on a system to identify one of ~50 known portrait backdrops in an image. I can generally isolate the background by cropping the upper corners of the test images (which tend to not contain portrait subjects) or use subject detection mask off areas that are not background.
I have a beginner's working knowledge of OpenCV and CV in general, mainly cascade training for object detection, and I'm looking for advice on the best strategy to classify test images according to background. Global characteristics such as hue, saturation, lightness, and std dev get me some of the way there. Would GLCM be useful to quantify texture? The backgrounds are occluded (but mainly abstract in image content--swirly backdrops and scenics), and appear at various scales in the test images.
Any pointers to papers, tutorials, suggested strategies will be most welcome. Thanks in advance.