Avian Visual Cognition

Dittrich & Lea Home Page

Next Section:
Motion Recognition

III. Motion Discrimination

This section summarizes some of the main findings about pigeons' ability to discriminate stimuli that involve some kind of movement (see also Lea & Dittrich, 1999). But since the ability to detect and to discriminate motion information depends on structural characteristics of the visual system, we will give first some basic details about birds', and in particular pigeons', eyes and the visual pathways to which they connect. A fuller treatment can be found in Husband & Shimizu's chapter of this volume. We also need to reflect a little on the characteristics of the apparatus that is commonly used to display moving stimuli, i.e. video and computer screens of various sorts. These are discussed more fully by D'Eath (1998) and Lea and Dittrich (1999).

Differences between bird and mammal visual systems

Butler and Hodos (1996) have reviewed the evolution and adaptation of neuroanatomical structures across the vertebrates, and from their highly informative book one can get an idea of the divergences between avian and mammalian visual systems. But of course we cannot really talk about a "typical" avian visual system. Like mammals, birds differ greatly in many aspects of their visual function, for example in the size and direction of their visual fields (Sillman, 1973; Martin, 1999), in their visual acuity, and in the dependence of visual acuity on light level are also wide (Martin, 1993). All the same, there are some consistent differences between the two groups, and some of these may have implications for movement perception. For example, because birds generally have flatter eyeballs than mammals, they have severe limitations on swiveling their eyeballs, so they rely on head movements instead to bring the important parts of the scene onto areas of its retina with high photoreceptor density. These head movements will necessarily also give motion parallax information, though of course it does not follow that the bird makes use of it. We suggest that in most non-flying situations the visual stability of the environment is not disrupted by birds' head movements leading to retinal change but, on the contrary, that head movements are compulsory to maintain visual stability. What appears at first glance paradoxically is a hypothesis that is firmly grounded on our notion of 'motion integration centers' (see above) and the structure of the birds' visual system. Unlike mammals, about half of the birds species that have been studied have more than one fovea on each retina. These dense cell areas are distributed in very different places in different bird species, leading to very different orienting head movements. Moreover, the cell density in extrafoveal areas of the retina is generally higher than in humans, as is the relative number of cones (see Butler & Hodos, 1996). 

These structural differences and variations do not make the study of birds' vision uninteresting to those whose primary interest is in human vision. On the contrary, a comparative approach offers an essential perspective. So long as it is based on biological principles, and not on mere botanizing, it is the one approach that can enable us to understand the evolutionary and ecological demands on visual perception and cognition (e.g. Kamil, 1988; Cook, 1993; Dittrich, Gilbert, Green, McGregor & Grewcock, 1993; Shettleworth, 1993; Lea & Dittrich, 1999).

Distinctive features of pigeon vision

Pigeons have a very wide visual field - of around 340°, compared to the 180°of humans. This is in common with most birds (though not all - consider owls) and many mammals (e.g. rats or rabbits). Given the high general cone density of birds, this means that in principle at least a pigeon can track an externally moving object without moving its head or eyes over much greater displacements than a human can. Pigeons are among the birds with a dual fovea (e.g. Hayes, Hodos, Holden,& Low, 1987; Husband & Shimizu, this volume (section IV)), and the two visual systems, frontal and lateral, seem to have rather different properties (e.g. Nye, 1973; Friedman, 1975; Martinoya et al., 1984); this means that an object that moves between them could pose problems for the bird's perception, unless there are mechanisms to compensate for the changes in the stimulation it will cause. 

Features of the motion displays

Video and computer screens have a number of well-known problems as stimulus generators for avian visual perception. They have limited resolution; they depend on apparent movement; their color generation is attuned to the human eye, not to the different photopigment set of the pigeon or any other species of bird, and in particular they emit no ultra-violet, which is involved in birds' color vision; they may flicker at rates that are perceptible to birds but not to the human experimenter; and of course they are inherently two-dimensional. We have explored these problems in detail elsewhere (Lea & Dittrich, 1999), and here simply state the conclusion we drew in that paper: that the video display is, if not a perfect tool, a highly acceptable tool for investigating birds' perception and cognition of visual movement. Birds may have particular difficulties in extracting from video displays the kind of fine detail required for discriminating between individual conspecifics, but many other kinds of discrimination can be made successfully. So in this chapter, we shall proceed without further comment to use such experiments to tell us what kinds of motion discriminations pigeons and other birds can make.

Is it moving or not and does it matter?

Two fundamental questions are whether birds can discriminate moving from static images, and whether they can generalize between moving and static versions of the same object. The first of these questions has been addressed by Siegel (1970), who showed that pigeons could discriminate simple static shapes from the same shapes in apparent movement at speeds of between 4 and 64 Hz. In further experiments, pigeons were able to discriminate between non-moving vertical lines, pulsating stimuli and those in apparent or real movement (Siegel, 1971). These findings can be seen as initial evidence that the positional change alone seems a salient feature used for discriminating simple stimuli. Hodos, Smith & Bonbright (1975) and Mulvanny (1978) investigated thresholds for movement detection, and found them to range from 4.1 to 6.1°/sec. However, these values are for frontal-viewing conditions and it can be expected that under lateral viewing conditions (e.g., optic flow) threshold levels may vary, as the lateral field seems to be adapted to faster moving stimuli. Threshold differences between different groups of birds depending on their ecology can also be expected. Dittrich and Lea (1993) extended the study of movement detection to much more complex movements, by training pigeons to discriminate between video scenes (of other pigeons performing characteristic responses) and still frames from the same scenes. 

Interestingly, there was a strong feature-positive effect in this experiment, with movement-positive discriminations being much easier to demonstrate than movement-negative discriminations with the same stimuli; this suggests, in line with the ecological data mentioned above, that movement is highly salient for pigeons just as it is for humans.  The converse question, of generalization from static to moving versions of the same stimuli, has also been addressed using category discrimination techniques. In an unpublished experiment, Watanabe, Lea, Ryan, and Ghosh trained pigeons to discriminate between a small set of pictures of birds and and a similar set of pictures of trees.  For example, one of the tree pictures showed an oak  and one of the bird pictures showed a pigeon (other examples included a gull, duck and  hawk, and other trees used included a fir, Japanese cherry, and scholar tree).  After training, tests were given in which the original training stimuli were presented in either smooth or staccato motion: the birds showed clear generalization of the discrimination to these new versions of the stimuli, showing that movement did not render the stimuli unrecognizable.

Oak tree stimulus as used by Watanabe, Lea, Ryan and Ghosh pigeon image as used by Watanabe, Lea, Ryan and Ghosh

To  see examples of the stimuli used by Watanabe, Lea, Ryan and Ghosh,  click on each of the above image to see them in smooth motion. To see the staccato motion conditions click here (example of tree, example of bird).

Discriminating between different kinds of movement

Simply recognizing whether or not something is moving is not a very interesting, however. It is more interesting to know whether birds can discriminate between different kinds of movement. Once again we have evidence from experiments both with simple and with more complex stimuli. As regards simple shapes, Siegel (1971) successfully trained pigeons to recognize the horizontal movements of approximately three vertical lines projected onto the response key. In a generalization test in which the horizontal movements were altered to vertical movements in steps of 22.5º pigeons showed a relatively steep generalization gradient as the direction of movement departed from the original horizontal movement displays. Furthermore, he also claimed perfect transfer between real and apparent movements in test trials. Pisacreta (1982) trained birds to track shapes moving, in a staccato fashion, at characteristic rates across a screen; changing the movement rate led to some generalization decrement, though not as much as changing color or form. 

A more complex case was studied by Emmerton (1986), in a rather more sophisticated investigation. She showed that when monochromatic stimuli were displayed on a fast oscilloscope screen, pigeons could be trained successfully to discriminate movement patterns. With the technology she used, there should be no uncontrolled problems with flicker, and color was irrelevant, so many of the difficulties of using video displays were overcome. The stimuli used were Lissajous figures, which are produced by moving a dot on an oscilloscope screen with independent sinusoidal motions in the vertical and horizontal dimensions; if the ratio of the frequencies is slightly different from an integer, the pattern generated changes constantly.  If the dot moves quickly, an outline shape is seen, though its form changes.  If the dot moves slowly, then a moving dot is seen, tracing the outline.  If the ratio of the frequencies of movement in the X and Y dimensions are changed, the shape generated by the moving dot changes. 

Emmerton (1986) Expt 3 resultsClick here to see the results from Experiment 3 of Emmerton (1986), in which pigeons discriminated between Lissajous figures generated by X:Y frequency ratios of approximately 2:1 and 1:1.  Of the four test types, "train" used the same stimuli and procedure as in training; "test" used the same stimuli but the reward contingencies were suspended; "nonrotate" again used the same stimuli as in training, while "rotate" used stimuli in which either the figure being traced was rotated through 90 degrees, or the direction of rotation of the dot was reversed. 

Interestingly, Emmerton showed that pigeons generalized to new views of outline figures but failed to generalize to views of dot-defined figures. Directional invariance seemed to hold only for a dot moving in a circle. She interpreted these test findings as evidence for stimulus-dependent generalization in pigeons. The only study which seems to have continued Emmerton's theme seems Cook and Katz's (1999) study on the perception of dynamic object properties. They trained four pigeons to discriminate computer-generated 3-D projections of cube and pyramidal objects. Generalization tests involved static and rotating presentations or modifications of these stimuli (for further details see Cook, this volume). It is interesting in the context of this chapter that they reported a dynamic superiority effect, i.e. dynamic conditions were always discriminated better than the static conditions (Cook & Katz, 1999, Fig.5). They interpreted this findus diagram and diagram of Emmerton results as indication that the objects' motion contributed information about their structure that was not obtainable from static presentations. These findings seem to strengthen the view that simple object motion can play a vital role in complex object recognition tasks. 

Although most of Emmerton's (1986) patterns were complex in themselves, the particular figures used as positive and negative stimuli were consistent from trial to trial, so the experiment was not a category discrimination in the way that so many movement discrimination tasks necessarily are. This difference can be highlighted when we compare pigeons' discrimination performance between artificial Lissajous figures and natural images. Pigeons successfully trained to discriminate between static and motion images of conspecifics showed remarkable response differences in generalizing between various types of movement (Dittrich & Lea, 1993, Figure 6). The authors reported a perfect negative correlation between the responses of the motion+ and the static+ group during generalization as well as a significant interaction between training group and type of motion displayed. Furthermore, discrimination was shown to be independent of size, perspective or viewing angle, brightness, and color. Both aspects of the findings were interpreted as evidence that pigeons not only can discriminate natural images on the basis of motion cues alone they also seem to form a concept of motion spontaneously. The next question to ask seemed at this stage whether pigeons can discriminate between exemplars of a motion concept and on what basis such a motion concept would be formed. Dittrich, Lea, Barrett and Gurr (1998) trained pigeons in a true concept discrimination based on movement types. In their Experiments 1 and 2, they demonstrated that pigeons could be trained to distinguish moving images of conspecifics from one another on the basis of the movement's category membership such as pecking or walking. Examples of the two ways of presenting the stimuli and the two types of movement presented are chosen from Dittrich et al (1998; click on images to see videos):

Category Membership
Full Detail
Point Light
Pigeon Pecking
Click here to see video
Click here to see video
Pigeon Walking
Click here to see video
Click here to see video
Several different scenes were used for each category, so presumably several different cues would be needed to make a reliable discrimination between the two sets (cf. Lea, 1984). Furthermore, in their Experiment 1 they used the standard 'pseudoconcept' control procedure: there was an additional group of birds for whom the positive and negative sets cut across the natural classification of the stimuli. The birds trained in this 'perverse pseudoconcept' task uniformly failed to discriminate, so we can assume that movements that belong to the same category to human eyes are also similar to one another from the pigeon's point of view. Lea et al. (in prep.) successfully trained bantam hens to discriminate the same set of video images of pigeons pecking and walking as Dittrich et al (1998) had used . This result (see section IV) shows, not just that the capacity to make discriminations between categories of movements is not confined to pigeons (which would have been very surprising), but also that it is not confined to images of conspecifics, which might be supposed to have some special status. In fact, as in experiments using color slide stimuli (Ryan & Lea, 1994), the chickens learned to discriminate pigeon stimuli rather faster than pigeons did.
Optic flow

As animals move around, most of the time they generate an informative pattern of optical changes. The optic array of natural surfaces, whether ground, sky, or water, registers a pattern of continuous changes with the animal's movement. This is called optic flow. For example, when moving forward or in parallel to or directly approaching a surface, there is a continuous flow of retinal projections relative to the animal's viewpoint. The information in this optic flow pattern is a reliable source for checking the relative velocity and the direction of movement. There is good evidence that birds possess the neurological apparatus to respond to optic flow variations (see, e.g., Wang & Frost 1992; Wylie, Linkenhoker,& Lau, 1997; Wylie, Bischof,& Frost, 1998), and some birds at least have a clear ecological need to do so (see, e.g., Lee & Reddish, 1981, who analyze the case of plunge-diving in gannets). Recently, Sun and Frost (1998) tested the response of neurons to the image expansion of objects approaching on a direct collision course with the bird. Three types of looming-sensitive neurons were described in the nucleus rotundus. Remarkably, none of these neurons seem to respond to a simulated approach of a stationary object. The authors suggest that these neurons provide precise information for the time-to-contact variable or the computation of tau. Similarly, behavioral experiments to link these two lines of research are obviously called for. 

Animals seem to have at least two methods for timing their reactions appropriately when embarked on a collision course either with an object or surface. First, they could estimate the time remaining before collision, or "time to contact": this could be derived from perception of the distance and speed of the movement of the object. However, neither of those would be directly available. Second, as Lee (1976) suggested, the ratio between retinal image size at a given instant and the rate of expansion of the image could be used. Lee gave this ratio the name tau. The ratio seems sufficient as long as the image size is not too large and the velocity of the approaching event relative to the animal is kept constant. Its value is independent of the speed and size of the object. It has also been reported that the landing responses of pigeons correlate well with tau (Lee, Davies, Green & van der Weel, 1993). Lee's time-to-collision ratio (tau) became a paradigm example of Gibson's ecological approach because tau requires no complex internal calculations in the brain and seems readily available from the flow field over the retina of each eye. So far, electrophysiological recordings (see above) from single neurons in the pigeon provide the best evidence for Lee's time-to-collision concept. Recently, it has been demonstrated that the information used in judging time to collision events seems of many origins (of which tau is just one) depending also, for example, on the task and stimulus condition (see e.g. Tresilian 1993; Wann 1996). Following our arguments on motion integrators one could say that not the detection of a single cue, namely tau, seems sufficient but the interactive processing of various motion parameters depending on the behavioral demands. More behavioral studies in this area are urgently needed to understand the role of motion variables in the determination of birds' reactions when approaching objects or surfaces.

Next section: Motion Recognition