| The analysis in this thesis is motivated by two models in the literature. Dong and Atick [4] describe the relationship between spatial and temporal frequencies of videos of natural scenes. Langer and Mann [10] describe a geometric constraint on the power spectrum of individual image sequences produced by an observer moving laterally relative to a cluttered 3D scene. Langer and Mann's model proposes a bowtie pattern in the frequency domain, but does not consider the distribution of power within this bowtie. In this thesis we consider the distribution of power within the bowtie by applying the model of Dong and Atick. The model predicts a spreading of power away from the bowtie axis, and a bias of power towards low speeds. Using synthetic image sequences, we verify both of these predictions. Other contributing factors are discussed including occlusion, transparency and the distribution of objects over different depths in the scene. |
