Often, differences in luminance, color, texture, and depth can help us determine object boundaries. However, when two surfaces have similar textures, as in the case of camouflage, or under dim lighting conditions, object segmentation can be difficult. In such cases, motion leading to the gradual accretion and deletion of texture information on a farther surface by a nearer one can be used to define the nearer object's boundary. It has been demonstrated that accretion and deletion is but one of a general class of texture element transformations that can give rise to the perception of illusory contours, global form, and global motion. This general process is called spatiotemporal boundary formation or SBF.
In the first chapter, I demonstrate two novel properties of SBF. First, SBF can be seen when element transformations are displacements in random directions. Second, global forms can be seen even when SBF-defined objects are rotating, expanding or contracting, accelerating, or smoothly deforming from frame to frame. I consider a two-stage model of SBF that can account for the perception of illusory contours and global form. In the first stage, oriented edge fragments are extracted locally from the sequential transformation of at least three elements in a small spatiotemporal neighborhood. In the second stage, these fragments are integrated and missing regions are interpolated by the same processes that govern spatiotemporal interpolation between contrast-defined edges.
Chapter 2 tests the first stage of this model. I created a display in which small circular elements were arranged in a sawtooth pattern and disappear and reappear one at a time in sequence. The resulting percept was not of apparent motion, but of an illusory bar that occluded elements one at a time. Using both subjective and objective methods, I identified the spatial and temporal parameters under which SBF occurs. The experiments provide support for models of SBF that begin with extraction of local edge fragments and identify minimal conditions required for this process.
In the final chapter, I implemented the first stage of the SBF model and used it to predict edge orientations of SBF-defined edges. Model and human performance were compared in an orientation discrimination task as a function of element density, number of element transformation, and frame duration. The ideal observer model was able to perfectly predict edge orientation while human performance was suboptimal. I considered several constraints and sources of noise that could contribute to differences between human and ideal performance. In a second experiment, I measured the sensitivity to spatial and temporal display properties that may have acted as sources of noise. A model that incorporated these constraints and sources of noise was able to model human performance very closely with no additional free parameters. The behavioral and modeling work provide the first empirical evidence in support of the two-stage model of SBF.
