Introduction

One of the first demonstrations of a motion-selective process was the waterfall illusion. After watching a cascading waterfall for a few seconds, Robert Addams shifted his gaze to the rocks on one side of the waterfall and was surprised to find that the stationary rocks appeared to move upwards, in the direction opposite to the water flow. This phenomenon, called the motion aftereffect, has been extensively studied. Prolonged viewing of motion in one direction (the adapting stimulus) desensitizes the observer to motion in that direction, such that a stationary stimulus appears to move in the opposite direction. Recent studies by Bob Sekuler and coworkers have shown that this desensi-tization occurs over a narrow range of directions centered on the direction of the adapting stimulus. These results suggest that the underlying motion sensors are direction selective and respond only to a narrow range of directions.

Another motion illusion is that stationary flashing lights on movie theater marquees appear to move. This phenomenon, called apparent motion, occurs when an object (such as a flashing light) appears briefly at one location, and after a short time delay a similar object appears at another location. In the case of theater marquees, the time and distance between adjacent flashing lights are large, so we can readily distinguish this motion from a real moving object. If the space and time intervals between successive presentations are small, apparent motion is indistinguishable from continuous motion. Although both television (NTSC 60 Hz or PAL 50 Hz) and movies (24 Hz) consist of animated sequences of static images, we generally perceive them as real motion. The perceptual equivalence of the real and sampled motions suggests that motion-selective neurons are insensitive to the additional information present in the continuous motion representation. In fact, the spatial and temporal sampling rates above which smooth and sampled motion are indistinguishable correspond to the human window of visibility—the limits of spatial and temporal frequency sensitivity of the human visual system (approximately 60 Hz and 50 cycles/degree).

Figure 1 provides a brief outline of the motion processing pathways in primates. It will serve as a road map as we discuss some of the important anatomical areas involved in motion processing. The pathway for visual motion processing in primates starts with the two main cell types in the retina: the magnocellular neurons that project mainly to areas involved with motion and depth processing and the parvocellular neurons that project mainly to areas involved with form and color processing. The magnocellular neurons project to two distinct layers in the lateral geniculate nucleus (LGN) in the thalamus on their way to cortex. From the LGN, these neurons project to layer 4ca in striate cortex, which is the first visual area in the brain (V1). About one-third of V1 neurons are selective for the direction of motion, but due to their small receptive fields they respond only to "local" motion. The V1 neurons in turn project to the middle temporal area (MT), which is an extrastriate area that is specialized for motion processing in primates. About 90% of the neurons in area MT are selective for the direction of motion; furthermore, these neurons are arranged in columns, according to the preferred directions of motion. There is a smooth gradation of preferred direction across neurons in neighboring columns. Neurons in area MT can detect combinations of local motions as well as the motion of a target relative to the background. The adjoining medial superior temporal area (MST) receives projections from area MT as well as from eye movement areas, such as the superior colliculus, the lateral intraparietal area, and the frontal eye fields. Area MST has neurons capable of analyzing

Figure 1 A broad outline of the brain pathways involved in processing visual motion information. Light is detected by the retina (top), which is connected to the lateral geniculate nucleus (LGN), which passes information to the visual cortex. Area V1 is the earliest cortical locus. From it, information is passed to the middle temporal area (MT) and the middle superior temporal area (MST), which are areas that specialize in the processing of motion information. The arrows show only the information going in the feedforward direction, but there are also feedback connections within cortical areas as well as from area V1 to the LGN. FST, fundus of the superior temporal area; VIP, ventral intraparietal area; LIP, lateral intraparietal area, FEF, frontal eye fields [modified with permission from Snowden, R. J. (1994). Motion processing in the primate cerebral cortex. In Visual Detection of Motion (A. T. Smith and R. J. Snowden, Eds.). Academic Press, London].

Figure 1 A broad outline of the brain pathways involved in processing visual motion information. Light is detected by the retina (top), which is connected to the lateral geniculate nucleus (LGN), which passes information to the visual cortex. Area V1 is the earliest cortical locus. From it, information is passed to the middle temporal area (MT) and the middle superior temporal area (MST), which are areas that specialize in the processing of motion information. The arrows show only the information going in the feedforward direction, but there are also feedback connections within cortical areas as well as from area V1 to the LGN. FST, fundus of the superior temporal area; VIP, ventral intraparietal area; LIP, lateral intraparietal area, FEF, frontal eye fields [modified with permission from Snowden, R. J. (1994). Motion processing in the primate cerebral cortex. In Visual Detection of Motion (A. T. Smith and R. J. Snowden, Eds.). Academic Press, London].

different patterns of optic flow and potentially estimating heading direction. This area also generates signals for smooth pursuit eye movements. During the course of this article, we will pose the problems that a generic motion analysis system must solve and present plausible candidate locations in the primate visual pathway. Although it is tempting to assume that a neuron with a particular property is responsible for the animal's sensitivity to that property, we believe that the animal's sensitivity is likely determined by a population of neurons sensitive to that property.

This article first discusses simple motion-related processing and then discusses more complicated motion phenomena. The organization of this article also reflects a progression from lower to higher visual areas. However, we must state that visual processing is not purely hierarchical; as evidence accumulates for certain interpretations of the visual scene, feedback from higher areas modulates the responses of lower areas, thus strengthening an emerging percept. The hardware for this interaction is certainly in place; there is abundant evidence for feedback connections from higher areas to lower areas. Also, perception might depend on more than the visual stimulus. The observer's past experience with visual stimuli can influence the perception and interpretation of a visual scene. For example, when a moving object disappears for a brief time behind an opaque occluder, there is no visual motion information during its traverse behind the occluder. However, we interpret the image correctly as an object moving smoothly behind an occluder rather than as an object that disappears for a brief time and coincidentally reappears with a similar trajectory at the other side of the occluder. This certainly has to do with our experience with the trajectories of moving objects and with occluders.

II. MEASUREMENT OF MOTION A. Receptive Field Structure

Several biological systems need to measure motion, whether it is a frog trying to catch a fly, a prey trying to avoid a predator, or a human trying to catch a baseball. Motion is the change in position of an object over time. Therefore, one way to measure motion is to identify an object, or a complex feature of an object, measure its location at one instant in time and again at another instant in time, and use these measurements to compute both the speed and direction of motion. Human visual motion sensors do not employ this strategy of detecting high-level features and keeping track of their locations over time, but instead compute motion signals by responding directly to local changes in intensity over time. Evidence from physiology shows that although the earliest neurons sensitive to motion respond to a wide range of stimuli, they respond best to a narrow range of spatial sizes and temporal intervals. The responses of most of these neurons can be described by simple receptive fields, which are oriented in space-time and are selective for both direction and speed. Many experiments have characterized these neurons' receptive fields by determining their responses to bars of varying widths, orientations, and speeds.

As visual information is passed from the retina to higher level processing areas, the receptive fields of motion-sensitive neurons become increasingly complex. Here, we focus on how the brain achieves motion-

sensitivity in the striate cortex. The responses of motion-sensitive striate cortex cells are directionally selective, which means that their response to moving stimuli depends on the direction of motion. Motionsensitive cells have large responses to motion in their preferred direction and much smaller responses to motion in the opposite direction. Direction selectivity requires receptive fields to be space-time oriented.

To appreciate how a space-time-oriented receptive field is built, we first consider the responses of nondirectionally selective neurons whose responses are independent of the direction of motion. These cells have simpler receptive fields that are space-time separable, which means that their space-time receptive field is the product of a spatial receptive field with a temporal weighting function. Readers are probably familiar with the concept of a spatial receptive field. The spatial profile of many V1 receptive fields is well characterized as a two-dimensional (2D) Gaussian, multiplied by a sinusoid as shown in Fig. 2A. This receptive field is localized in space and responds best to a particular orientation. The cell's preferred orientation and width are determined by the orientation and spatial frequency of the sinusoid. Neural responses to stimuli also depend on time. The neural response to a brief flash of light extends over a period of time (about 100 msec) and is described by a temporal weighting function (or temporal impulse response function). Space-time-separable receptive fields are the product of this spatial receptive field with a temporal weighting function. The temporal weighting function is generally modeled as a biphasic response with a positive response that is followed by a more extended negative response. Two examples of temporal response functions are shown in Fig. 2B. For space-time-separable receptive fields, the spatial receptive field polarity changes with time, but its profile does not change (i.e., it is not oriented along the time dimension). Figure 2C shows an x-t plot of a typical space-time-separable V1 receptive field that is selective for a static vertical bar. This space-time-separable receptive field is unselective for motion direction (i.e., it responds similarly to leftward and rightward motion). This can be seen by comparing such a receptive field profile to Fig. 3, which shows x-t plots of a bar that is moving rightward (Fig. 3B), moving leftward (Fig. 3C), or is static (Fig. 3D).

In contrast, a motion-sensitive cell responds strongly to motion in its preferred direction but weakly or not at all to motion in the opposite direction. Figure 4B shows four examples of motion-sensitive receptive fields that are oriented in space-time; their response at each spatial location depends on time. Such a neuron is selective for a particular velocity and direction. Achieving a spatial response that changes with time is difficult because stimulation of a particular receptive field region must produce a positive, negative, or zero response depending on the time. One way to construct a space-time-oriented receptive field is to combine space-time-separable receptive fields (Fig. 4A shows four space-time-separable receptive fields). Space-time-oriented receptive fields can be constructed by combining (either adding or subtracting) the outputs of two such cells, such that the response of the second is shifted in space and time relative to the first (Fig. 4B shows four different space-time-oriented receptive fields constructed by combining pairs of the receptive fields shown in Fig. 4A). Space-time-oriented receptive fields are selective for motion direction, but their responses are polarity specific; their response to a dark bar is inverted compared to their response to a bright bar. To detect the motion of an object reliably, the neuron's response should not depend on the polarity of the moving bar. One way of achieving this is to square the response. It appears that the visual system achieves a polarity-insensitive filter by taking the sum of the squared responses of the two filters shown at the top (or bottom) of Fig. 4B. This measure based on the squared responses of local direction- and speed-selective units is called motion energy. Ted Adelson and Jim Bergen proposed that this motion energy stage is followed by an opponent stage that takes the difference of local detectors tuned to opposite directions (e.g., the difference between rightward and leftward responses). Physiological measurements support the existence of space-time-oriented receptive fields that compute motion energy, but they do not support the existence of an opponent energy stage.

Unraveling Alzheimers Disease

Unraveling Alzheimers Disease

I leave absolutely nothing out! Everything that I learned about Alzheimer’s I share with you. This is the most comprehensive report on Alzheimer’s you will ever read. No stone is left unturned in this comprehensive report.

Get My Free Ebook


Post a comment