Classify new test images

1. For each patch in the epitome, determine whether or not it maps into the test image.
2. Use Bayes Rule and the patch mapping conditional probabilities to calculate the odds ratio of
   

   $$\frac{P(pos\vert patches)}{P(neg\vert patches)} = \frac{\prod_i{P(patch_i\vert pos)}}{\prod_i{P(patch_i\vert neg)}}$$ (3)

   
   
   where $i$ ranges over the indices of all patches being used to classify.

3. Classify the test image using some threshold for the odds ratio.

The general classification decision framework and the Bayesian odds ratio calculation are very similar to those used in [2].