Images of objects contain information at many spatial scales. The experiments described in this dissertation investigate whether integration of information at different spatial scales, in human object recognition, satisfies the two requirements of a linear system, superposition and homogeneity. If a system satisfies the principle of superposition, the system's response to two stimuli presented together is the same as the sum of the system's responses to the individual stimuli presented separately. Such a model can be called "additive". On the other hand, interactive models of object recognition propose that information at one spatial scale influences the value of information at another spatial scale. For example, information relevant to coarse form may aid the interpretation of local details. We tested these alternatives by measuring the usefulness of low spatial frequency (coarse) and high spatial frequency (fine) information during the timecourse of object recognition, in both a between-subject and a within-subject paradigm. We used the data to test both a probability summation model and a linear model of information combination. Both models fit the data reasonably, but overall the linear model proved to fit the data better than the probability summation model. If a system satisfies the principle of homogeneity, then if input to the system is multiplied by a constant factor, the output will be multiplied by that factor as well. We were able to use one set of parameters to model subjects' performance in two experiments, one that used stimuli with half the contrast of the stimuli used in the other. We interpreted this success as support for homogeneity, the second requirement of a linear system, in spatial scale integration in object recognition. These results imply that information of different spatial scales is combined linearly in object recognition, rather than being processed by separate spatial frequency channels that combine outputs by probability summation.
展开▼
