Avian Visual Cognition

Dittrich & Lea Home Page

Next Section:
Motion Discrimination

II. Retinal motion and motion perception

There are four different reasons why the image of an external object might move on an animal's retina.

1. The object has moved 
2. The animal has moved its location
3. The animal has moved its head relative to other parts of its body
4. The animal has moved its eyes relative to its head 

Obviously, two or more of these could happen at the same time; and by coincidence or by design, they might even cancel each other's effects out so that no retinal movement results. Equally

obviously, the ecological implications of the four different types of movement are very different. All too often, we just discuss movement perception as though all we had to consider was the perception of a moving object by a static subject. That is almost never the only thing that is happening, and in the case of birds (probably the fastest moving class of vertebrates, on average) it is certainly not the only important case. Therefore, in this chapter we give the discussion of motion detection and recognition in pigeons a broader perspective, but first discussing studies on the relationship of the birds' own movements to the changing or moving image on the birds' retina (retinal motion). The processing of retinal motion information may be quite different depending on whether, and how, the bird itself is moving, and it is the job for the behavioral scientist to separate these different aspects of motion processing.

Visual stability while moving

Animals are always moving around, and most retinal image motion is caused at least in part by movement of the animal itself. There are, of course, compensatory eye movements that can reduce the effects of retinal movement but they can never remove it completely. For example, during pecking movements, when the head moves downwards, it seems that pigeons are able to make convergent eye movements to keep the view of seeds in focus (e.g. Martinoya, Le Hourezec & Bloch, 1984). Therefore, pigeons must be able to distinguish between self- and externally-generated retinal motion if they are to perceive and use visual motion information appropriately. A moving pigeon seems to have three sources of information available about self motion: a) visual consequences, b) motor commands, and c) proprioceptive feedback. The task for the behaviorist and the physiologist is to find out which information is actually used under which circumstances. For example, the retinal motion information might be supplemented by proprioceptive information about the position of the head with respect to the surrounding space, of the head with respect to the body, of the body with respect to the surrounding space, of the eyes with respect to the world, and the eyes with respect to the surrounding space. In order to make distinctions between self- and externally generated retinal motion possible, it has been assumed  that an animal's movement produces information that is made available as feedback into the visual system. Such 'reafferent' processes were postulated and discussed by Erich von Holst and Horst Mittelstaedt who published an extremely influential paper in 1950 under the title "The reafference principle: Interaction between the central nervous system and the periphery" (translated from the German). 

One assumption of the reafference principle is that if the animal moves its eyes or head, the stimulation generated by the change of the retinal image constitutes a reafference. The stimulation is called reafference because this information is the product of its own movement. It is distinguished from a motion stimulation as a consequence of an object moving. Therefore, in some way the animal distinguishes between movement of objects relative to its eyes, when this is a consequence of its own movement, from the relative movement as a result of objects moving externally.  It is hypothesized that self-initiated movement is triggered by a command (="move" or movement) from a higher center which sets up an 'efference copy' or 'corollary discharge'  in a lower center. This may then send motor impulses or 'efference' to the muscles. As a result of the movement a 'reafference' stimulation, e.g. actual retinal motion,  is compared with the efference copy in the lower center, e.g. anticipated retinal motion. Visual stability, for example, is achieved if the two signals compensate each other. Otherwise, orientational movements to compensate for the discrepancy between actual and anticipated retinal motion would follow.  Independent of v.Holst and Mittelstaedt's (1950) report that the fly Eristalis turned indefinitely when its head was turned by 180 degrees, Sperry (1950) reported similar observations in frogs and fishes. In these experiments the orientational coordinates of the visual system were manipulated and as a consequence the actual visual input and the anticipated visual input seemed always incompatible, leading to some form of compensatory movements indeed, e.g. circling.

Although the reafference principle has been more generally applied to kinesthetic stimulation and animal's movement in space (see v.Holst 1954) - including our own movements, its relationship to a more complete conception of the visual organization of the surrounding space and its cognitive implications have yet to be fully explored. Furthermore, it can be assumed that the different visual systems of pigeons, such as frontal visual system and lateral visual system, might play a very different role in processing retinal motion and that feedback about the magnetic orientation as well as the polarization of the light rays might also be processed in this respect. This long list of potential sources of information reminds us that the problem of allocating retinal motion to self motion or object motion is an extremely important one, ecologically speaking. The fact that so many systems contribute to the solution of this problem is evidence of both importance and its complexity. Here, the view is forwarded that visual stability during movement is a result of the multiple interactions of visual, efference copy and proprioceptive as well as cognitive information (see Mergner & Rosemeier, 1998). As important as the notion of 'efference copy' appears to be, this principle on its own does not seem to provide a sufficient explanation of visual stability in all cases.

One of the open questions, of course, is the problem of the appropriate coordinate system both for representing the space surrounding an animal, and at the same time for guiding behavior. Such a coordinate system requires a spatial reference point and procedures for spatial referencing, e.g. a geometry. Possible reference frames could be the retina, the eyes, the head, the body or some areas of extra-body space. Possible geometries include the Euclidean system of three orthogonal axes, scaling the X, Y and Z dimension. However, in the context of human visual perception such geometries have been found arbitrary and biologically implausible (e.g. Koenderink, 1986, Simpson, Leonard & Soodak, 1988), and these problems would even be more severe in the case of a flying bird or a piscivorous bird with the eyes above water to a submerged prey with its bill. The visual mechanisms and spatial geometries of prey capture in water birds are further discussed in Katzir (1993) and Lythgoe (1979). 

In a biological context, the effective axes of spatial referencing seem to be non-linear and non-orthogonal, and several different solutions might be used, as long as spatial representations can be converted from one coordinate system into another. For example, to convert from representing space based on retinal coordinates into a body-referenced system requires information about the position of the head with respect to the body. Such conversion is not necessarily simple and may require additional information. Normally, the motion of organisms or their parts in space can be characterized by a combination of translation and rotation. The speed of translation refers to the rate of change of distance (e.g. meters/second) and the speed of rotation refers to the rate of change of angle (e.g. degrees/second). The same characterization of motion can be used when referring to moving object images on the retina. There is no single "correct" solution for spatial referencing as long as conversion from one into another reference frame is possible, for example from retinal coordinates into extra-body coordinates. A convenient method for describing motion characteristics is called  vectorial representation and tries to avoid some shortcomings of the orthogonal or linear reference systems. The  method of vectorial representation seems equally useful when describing either the rotation of the head or body, or when describing its visual consequences. In either case the result is a set of retinal flow fields that can be described in vector terms.  Martin Hoffman (Australian National University) has compiled a neat self-motion demonstration of what this looks like:

The distinction between self motion and object motion has important consequences for the organism. Therefore, it has to be robust: it must work under a wide range of circumstances and at all times. The pigeon that mistakes its own movement for a movement of the building it is flying towards is in for a rude shock. As Walls (1942, p.343) put it: "No sense other than vision is at all reliable for the orientation of animals with respect to the objects in space.  And the big reason why it is vital to know where things are is that some of those things, and the animal itself, move."

More recent discoveries about, for example, echo-location might cause us to qualify the first clause.  At least two groups of birds are now known to include some echo-locators  (swiftlets, see for example Griffin and Thompson, 1982, and oilbirds, see e.g. Suthers & Hector, 1985).  But these remain exceptions, and in any case Walls's second clause remains powerfully persuasive: position has to be constantly monitored just because it changes.

In the case of object motion, movements of the retinal image may in many cases provide the features necessary to determine the structure of the surrounding space or the 3-D structure of objects. Walls argued that one could not imagine vision without the capacity for motion processing. Other abilities, such as high levels of spatial resolution, or color processing, might be dispensable, and present in one species but not of another; but a species that could not "make sense" out of retinal motion could barely be said to see at all. In his rather old-fashioned terminology, Walls writes of the great importance of the "awareness of motion". Interestingly, he also pointed out that sessile organisms, plants or even relatives of eyed animals like sealilies and barnacles, though they lack eyes altogether, may still be able to respond to movement. The shell-closure reflex of barnacles is a fine example: here the approach of a potential predator that casts its shadow on the few photosensitive receptors, by measuring changes in light intensity, highlighting the role of motion from an evolutionary viewpoint at an early stage. A related point is that neurons sensitive to directional motion are surprisingly common in the retina and optic tectum of lower vertebrates, especially amphibians and fish (e.g. Ewert, 1982;  O'Benar, 1976), compared with the situation in cats and monkeys. It seems that the most basic (and therefore evolutionarily most essential) kinds of visual processing involve mechanisms of motion detection.

Several studies of the optic tectum in birds, mainly of the pigeon, show that a large proportion of the neurons are motion-sensitive with a center-surround organization and therefore selective for relative motion (e.g. Jassik-Gerschenfeld & Guichard, 1972; Frost & DiFranco, 1976). However, to construct the structure of the surrounding space and objects from a moving retinal image, the organism seems to need some knowledge about the 3-D structure of the world including objects. It is at this end of the process where the study of animal cognition from an evolutionary-ecological viewpoint might be able to make a contribution, in revealing the properties of the pigeon's knowledge about the world, and gaining an understanding of the contribution to it of the perception of moving objects.

    The role of head movements: The "visual input enhancing" hypothesis

For example, it seems reasonable to assume that there would be a strong relationship between a bird's typical head movements and its visual processing. A common form of head movements in pigeons can be called head bobbing and has been described as a thrust-hold cycle (e.g. Friedman, 1975; Frost, 1978). During the thrust phase the head is rapidly thrust forward, whereas during the hold phase it seems as it remains virtually motionless with respect to forward or backward movement (our clear visual impression that pigeons retract the head at some point during the movement is an illusion). Head bobbing movements are widespread among bird species, being found at least in 8 out of the 27 families of birds one can find representatives of head bobbing (Frost 1978); Dagg (1977) lists 28 species that demonstrate head bobbing during locomotion. There are at least three ways in which head bobbing might be related to visual processing and, surely, the different hypotheses proposed should not be seen as exclusive:

1. The head movements may allow the birds to process some kind of visual input which otherwise the birds would either not be able to sense or process. In this case, head movements would generate some kind of relative motion which is necessary if the birds is to detect or recognize objects or their characteristics. We argue that in this case head movements do not lead to the interruption of  visual processing in some way but, on the contrary, such head movements seem to be  necessary to guarantee appropriate visual processing. Therefore, we hypothesize that head bobbing movements can provide opportunities for enhanced visual processing. We can call this the "visual input enhancing" hypothesis of head bobbing. 

2. Even if head bobbing has developed independently of any immediate role in the visual processing of object information, the relative motion images generated on the retina would still result in specific processing requirements for the visual system - not least because the birds have to distinguish between self-generated relative motion on the retina and externally-generated flow patterns on the retina. We can call this the "gaze stabilizing" hypothesis of head bobbing, as the bird has to compensate for the retinal flow in order to maintain visual stability. Recently, it has been shown by using videography that during the hold phase the head of pigeons is really held stable in space (Troje & Frost, 2000). In this study the pigeon's head-bobbing has been interpreted as an optokinetic response to stabilize the retinal image during the hold phase. 

3. A different kind of compensation process would be involved when head movements are not shown in form of the typical head bobbing rhythm, but in the form of adjusting head movements or some form of eye movements (e.g. Fox 1978, Erichsen et al. 1989, Wohlschlager, Jager & Delius, 1993) that actually serve to maintain a more constant retinal image. We can call this the "visual compensation" hypothesis. One way of compensating for retinal flow would be to suppress the processing of visual input during head or eye movement altogether. In humans, this is known as 'saccadic suppression' and a similar mechanism has been postulated in pigeons. Pratt (1982) found that saccades were made during 80% of the thrusts of head bobbing, and not made in the hold phase. This result leads to the assumption that during thrusts with saccades there is probably no visual input because retinal motion information is suppressed.

The 'gaze stabilizing' and the 'visual compensation' hypotheses both seem to strongly imply a need to integrate signals from the vestibular system, indicating head/body movements, with signals from the visual system, indicating optical flow signals on the retina. Such an integration of the two signals might result in a body velocity signal that reliably indicated the spatial relationship of the bird to its environment. Precise information about the bird's own spatial position in relation to its surrounding is crucial for any visual processing of motion information. A nice example for the relationship of head movements, locomotion and visual perception of space can be found when chickens have to jump over gaps of different depths and widths, walking on different surfaces and facing a visual cliff (e.g. Green 1998a, Green 1998b). Head orientation was defined by the angle between the horizontal and the line linking center of the eye and the tip of the beak. The orientation was measured at the point where a jump began based on video recordings. It was found that up to one-week old chicks showed a consistent upward rotation of the head if the gap became wider or if the far side was raised. According to Green head orientation seemed independent of the fixation of the far edge at a particular elevation in the visual field, that means any hypotheses based on specialized retinal areas or visual field structure (e.g. myopia) might well be inappropriate. Instead, there was a consistent relationship between head orientation and the initial trajectory of the jump movements; each 1 degree change in trajectory change for a potential jump was accompanied by an approximately 1 degree change in head orientation. In such situations as jumping or landing the relationship of head/eye movements and visual information appears crucial for birds in controlling such parameters as body speed and direction, foot extension during jumping and landing. But also during walking, as Green found,  head orientation seems closely correlated with the slope of the surface, downward slope leads to an increase in head angle and vice versa. Interestingly, the same effect was found when the animals walked on a level transparent surface above a slope, ruling out a  merely simple kinaesthetic or motor interpretation. Therefore, here it is suggested that head movements play a crucial role not only in gaze stabilization but often more unexpectedly in enhancing visual input.

Studying the role of head and eye movements in pigeons Wohlschlager et al (1993) simultaneously recorded head and eye movements during locomotory and pecking behavior. The pigeons were trained to traverse a conditioning chamber, with a pecking key and a food dispenser at each end. Each trial involved key pecking, walking, and feeding. Head movements were registered with a skull-mounted miniature accelerometer, and eye movements were recorded with implanted electrooculogram (EOG) electrodes. They found an almost perfect temporal coordination between head and eye movements during both walking and feeding bouts. The findings have been interpreted that during walking, head movements primarily provide retinal image stability, and eye movements support visual scanning. Then, during feeding, head movements mainly subserve the grasping of food items, and eye movements maintain visual fixation on them. These findings illustrate the 'visual compensation' hypothesis of eye or head movements. 

Of course, birds' movement perception has been studied in relation to visual flow fields. But it seems that, in the past, birds' performance has not been systematically linked to their own movements during perception. In other words, the perceptual and cognitive consequences of the birds' own movements when being tested have not sufficiently been taken into account when discussing motion perception. In this context, it is interesting that Frost & Nakayama (1983) have described neurons of the pigeon optic tectum which are inhibited by in-phase motion of a small object against a background but facilitated by anti-phase motion. The inhibitory surround seems very large and responses seem only broadly tuned to motion direction. These are the characteristics we would expect if these neurons are involved in detecting relative motion of objects triggered by the bird's own motion. But it would be a poor motion detection system that could only tell self-motion from object-motion by the proportion of the visual field that moved. Obviously a much better job can be done if objects in the world can be identified, and classified as likely or unlikely to be in motion. Conversely, the characteristic modes of motion of an object must have an important role to play in the identification of that. Thus the remainder of this chapter concentrates on the interaction between movement perception in birds (mainly pigeons) and the concepts through which they categorize the objects in their visual world. A highly interactive process seems to be required if an organism is to simultaneously interpret self motion information including retinal motion, on the one hand, and categorize objects, on the other. But in the case of humans' sophisticated ability to perceive the movements of others it has been argued that it is exactly this interactive nature of the processing mechanisms which contribute to many of the astonishing results in motion perception, outlined in Dittrich's (1999) 'interactive encoding model' of movement perception. Following this line of argument, we suggest replacing the idea of 'motion detectors' with the broader notion of 'motion integrators' to capture the cooperative, flexible and adaptive nature of such motion processing mechanisms, particularly in vertebrates (for a discussion see Dittrich 1999). On this view, it is an important task both for the behaviorist and physiologist to determine the working principles of the motion integration centers responsible for the interactive processing of various sources of motion information. The following sections of the chapter give an overview of the contribution to this task from current work on avian cognition and behavior.

Next section: Motion Discrimination