Moving from spatially segregated to transparent motion: a modelling approach

Abstract

Motion transparency, in which patterns of moving elements group together to give the impression of lacy overlapping surfaces, provides an important challenge to models of motion perception. It has been suggested that we perceive transparent motion when the shape of the velocity histogram of the stimulus is bimodal. To investigate this further, random-dot kinematogram motion sequences were created to simulate segregated (perceptually spatially separated) and transparent (perceptually overlapping) motion. The motion sequences were analysed using the multi-channel gradient model (McGM) to obtain the speed and direction at every pixel of each frame of the motion sequences. The velocity histograms obtained were found to be quantitatively similar and all were bimodal. However, the spatial and temporal properties of the velocity field differed between segregated and transparent stimuli. Transparent stimuli produced patches of rightward and leftward motion that varied in location over time. This demonstrates that we can successfully differentiate between these two types of motion on the basis of the time varying local velocity field. However, the percept of motion transparency cannot be based simply on the presence of a bimodal velocity histogram.

A model of neural mechanisms in monocular transparent motion perception

“…These findings reveal important parameters of tuning properties of the basic mechanisms for motion detection. Recent work by [7] have shown, however, that multimodal velocity distributions (histograms) are not sufficient to indicate transparency percepts (compare also our first computational study depicted in Fig.2). We have emphasized how different motion directions can be segregated computationally based on generic neural interactions using the neural representation of velocities in area MT.…”

“…The segregation of transparent motion patterns has been investigated psychophysically by 2 For random dot kinematograms used as input only 3 speed values were employed, and parameters of the center-surround mechanism were adapted to this new configuration ('σ using stimuli consisting of stripes of a given width which contain coherent movements where direction alternates between pairs of stripes [12,5,7]. Stripe widths in the employed random dot kinematograms varied from 8 to 64 px.…”

“…As a main contribution, in order to enable the perception of transparent motion, it is proposed how multiple velocities in a log-polar sampled speed-direction space can be represented at individual spatial locations and how the neural interactions are defined. Other computational approaches, including computer vision algorithms, attempt to optimize representations with respect to single velocities and investigate the necessary conditions in representing multiple motions corresponding with transparent layers [20,7,23]. Here, we suggest local competitive centersurround interaction between responses in velocity space that leads to mutual competition or coexistence of multiple motion representatives depending on their separation in velocity space.…”

Asymmetric global motion integration in drifting Gabor arrays

“…For rigid frontoparallel translations, the motion constraints depend solely on the speed and direction of the 1-D velocity components and are independent of location in the visual field (Jasinschi, Rosenfeld, & Sumi, 1992;Schunck, 1989), i.e., we can throw away position information and still compute the global motion. However, for more complex situations, such as multiple objects, global rotations and expansions, and transparent motion, the spatial arrangement of motion signals must be taken into account (Durant, Donoso-Barrera, Tan, & Johnston, 2006;Zemel & Sejnowski, 1998). Accordingly, a number of motion-processing models make explicit use of the spatial distribution of motion signals (Grossberg, Mingolla, & Viswanathan, 2001;Liden & Pack, 1999).…”

Real time stereo-based biologically inspired 3D motion classifier cells

“…On the other hand, one of the fundamental problems in the neural computation of 3D motion is the grouping of velocity signals into surfaces as in the case of motion transparency [5], where local moving elements appear to be grouped into two or more spatially overlapping surfaces. Hence, the challenge for modeling 3D motion transparency is raised in order to demonstrate how two different motion signals can appear perceptually co-localized in the same space.…”