“…As can be seen, features obtained from the Intensity Ranklets alone perform better than those obtained from the Variance Ranklets alone. This is in accordance with the findings of psycho-physical studies on the relative importance of the two types of stimuli the filters are designed to detect [9]. The best results are however obtained by combining the two sets of features: the addition of second order information more than halves D1, D3, D4, D5, D6, D8, D9, D11, D16, D17, D18, D20, D21, D25, D26, D27, D28, D29, D33, D36, D78, D87, D92, D95, D101, D102 4 D1, D3, D6, D11, D16, D17, D20, D21, D24, D28, D29, D32, D34, D35, D46, D47, D49, D51, D52, D53, D55, D56, D57, D65, D78, D82, D84, D85, D101, D104 the error rate on three of the datasets.…”