Deficits in the Perceptionof Discontinuities from Motion Cues
(clicking on the reference numbers in thetext below will bring you to the reference and abstract)
Double dissociation of deficits. In 18 patients with unilateral brainlesions we demonstrated a double dissociation of deficits on motion coherenceand motion discontinuity detection in random dot kinematograms (RDKs).The stimuli consisted of sparse RDKs as illustrated in panels e) and i)of the Core Motion Tests. Thisresults provides further support to our previously reported findings intwo patients which suggested that, contrary to several popular computationaltheories of motion discontinuity, coherence (which requires global integration)and discontinuity are not computed simultaneously (nor is discontinuitycomputed at a stage that follows coherence computation). In 4 patientswith bilateral lesions  and in 12 additionalpatients with unilateral lesions in the dorsal pathway, we found a similardissociation when the stimuli embodying the motion discontinuity consistedof dense RDKs.
Acomputational model for motion discontinuity. We (Vaina and Sundreaswaran)developed and implemented a local model that explains the patients’ dataand tested the model with the same stimuli that were used in the psychophysicalstudies of patients’ performance on these tasks. The goal of the localmodel was to detect motion discontinuity based on measurements within asmall aperture. Our model extends Nakayama&Loomis’ model for the situationwhen only a local projection of the optic flow is available (due to theaperture problem). We used a voting scheme where each normal flow votesa set of directions and the resulting approximation to the convexity functionof Nakaayma&Loomis is thresholded to find locations with significantmotion discontinuity.
The local projection, termed normal flow,can be computed easily from two or more frames of an image sequence. Theaperture problem posits that only the component of optic flow along thelocal intensity gradient direction can be computed. To address this informationloss, we extended the Nakayama-Loomis model in the following way. For eachdirection considered in the Nakayama-Loomis model the projection of opticflow vector in that direction was used in calculating the value of theconvexity function. Since our model did not compute optic flow, we deviseda voting scheme so that every normal flow vector voted for a set of directions.
Without knowing which is the correct opticflow vector, it is impossible to calculate the magnitude of the components.Given this limitation, we treated all components equally hence every componentwithin a 90° range of the normal flow vector gets a vote. To computethe convexity function for each direction di we added the votes withinthe central region C (NiC) and subtracted k times the total votes withinthe surrounding region S (NiS). After squaring this difference, we summedthe result for all directions (eq. below).
The resulting approximation to the convexityfunction of Nakayama and Loomis was thresholded to find locations withsignificant motion discontinuities. On a set of dense patterns we showedthat the model computes discontinuity, and thus motion discontinuity canbe detected by using local motion processes. We suggest that such localcomputation may be used by the visual system when the global motion mechansimsare impaire, or even when the local motion cue is strong and available.
1. Vaina LM, Gryzwacz NM, LeMay M, BienfangD, Wolpow E “Perception of Motion Discontinuities in Patients with SelectiveMotion Deficits” in High-level motion processing: computational, neurobiological,and psychophysical perspectives, T. Watanabe (ed), MIT Press, 1998; pp.213-247.