Feature blindness: A challenge for understanding and modelling visual object recognition
Fig 2
An illustration of the four types of test conditions.
Each category has two diagnostic features: here, the overall shape and the colour of one of the segments. In the training images, features are mapped to categories using the following mapping: {(Shape A, Red) → Category 1; (Shape B, Blue) → Category 2}, where Shape A and Shape B are the shapes on the left and right, respectively. In the Both test condition, both types of features (shape and colour) have the same mapping as training. In the Conflict condition the mapping of the non-shape feature is swapped—i.e., the new mapping is {(Shape A, Blue) → Category 1; (Shape B, Red) → Category 2}. In the Shape condition, images have only one diagnostic feature—the overall shape—which has the same mapping as training: {Shape A → Category 1; Shape B → Category 2}. In the Non-shape condition, images have no coherent shape, but contain the same diagnostic colours as the training images: {Red → Category 1; Blue → Category 2}.