Abstract
A fundamental question in vision science is whether objects can be understood in terms of their parts. If the whole is the sum of parts, it suggests that dissimilarities between whole objects can be explained using dissimilarities between their constituent parts. Here we set out to investigate this problem by studying dissimilarity relations in a set of abstract objects. We measured perceived dissimilarity by asking subjects to find the oddball item in a visual search array. The reciprocal of the time taken by humans to find one object among homogeneous distracters was taken as a measure of dissimilarity between the two objects. In Experiment 1, we created two-part objects joined by a stem. Each side of the stem could be occupied by one of seven parts, resulting in a total of 49 objects. Subjects performed searches involving all possible 1,176 pairs (49C2) of these objects. We then asked whether these search dissimilarities could be explained using a smaller set of 21 (7C2) part-part dissimilarities. Our main finding is that dissimilarities between a pair of objects is a linear sum of the dissimilarities between every pair of their constituent parts (r = 0.88, p < 0.0005). Thus, for objects AB and CD, their overall dissimilarity is a linear sum of the dissimilarities between AC, BD, AD, BC, AB and CD. In Experiment 2, we confirmed that this result holds for three-part objects (r = 0.90, p < 0.0005). The only systematic deviation between observed and predicted dissimilarities was for searches involving symmetric objects. For pairs of symmetric objects, the predicted and observed dissimilarities were tightly correlated but offset by a fixed amount. This suggests that symmetry confers an additional but fixed distinctiveness to objects. Taken together, our findings reveal a surprising and systematic linear rule by which objects are related to their parts.
Meeting abstract presented at VSS 2015