EVALUATION PARADIGM FOR SCENIC ESTIMATION PROBLEM

• 637 images and their corresponding subjective SBE ratings

  
  
  

• Divide the database into train and test sets each with similar distribution as the original database (4 cuts of train-test pairs)

  
  

• Features are extracted from each image (i. e. color, lines, fractal)

  
  

• Build different models using different combinations of the features

  
  

• Evaluate these models on the test data and compare the performance of these models against the reference model (subjective SBEs)