Excerpt from
L.Kleine-Horst: Empiristic theory of visual gestalt perception. Hierarchy and interactions of visual functions (ETVG), Part 10,V

Comparing the ETVG with the Grossberg theory

1. Similarities

a) Rough structural comparison

Because two readers had recognized some similarities between the psychological ETVG and Grossberg's neurophysiological theory, the later so-called FACADE theory, I shall compare these theories with each other to a certain extent, and try to point out a number of similarities and differences between the ETVG and both the first Grossberg theory (Grossberg and Mingolla, 1985a, pp.141-171, 1985b, pp.172-211; 1987, pp.116-125) and his second theory (Grossberg 1993, pp.463-483; 1997, pp.618-658; Grossberg and Pessoa 1998, pp.2657-2684). (The reference to page numbers is sufficient to know which article is referred to; italics and quotation marks in quotations will be as in the original.)

Both the Grossberg theory and the ETVG started with the fact that the visual system does not precisely transform stimulations into experience; Grossberg mentioned the lack of retinal stimulation at scotomas and the blind spot, and the "discounted illuminants" caused by veins that occlude photoreceptors. This deficit, however, does not reduce the experience of both boundary and "features" (i.e. brightnesses and colors) projected onto these regions. To account for these facts, and others such as subjective contours, occluded figures and neon color spread, the author assumes two different, parallel systems to exist:

In the second Grossberg theory, mutual influences also between FCS and BCS and between both FCS and BCS and the Object Recognition System (ORS) are proposed.

Figure 10-4. Monocular perception system according to the ETVG (A) and the FACADE theory (B)

ETVG: The ETVG could agree with the hypothesis that the processes corresponding to MP, BCS, and FCS may be "preattentive", and the ORS capable of supplying the perception with memory contents (explicitly acquired, in this case). MP seems to correspond to the totality of the ETVG gestalt factors Dm and Dl, and their lower-level factors Pml, Zm, and Zl.

In order to compare the structures of the theories, in Fig. 10-4 the ETVG visual factors from Zl to Fl, responsible for the static, two-dimensional figure/outfield perception, as well as the hierarchical position of higher factors, are depicted twice. In A, the factor symbols are connected by two-way arrows in order to show which immediate interrelationships are established between the factors and their antagonistic positive/negative functions, according to the ETVG. (Fig.10-4 corresponds to Fig.2-14, but has been expanded to show the antagonistic relationships.)

When comparing two theories with each other, one often encounters difficulties founded on the use of different languages and terminologies. Hence, I shall try to translate - at least for myself - the Grossberg language into the ETVG language. There are, of course, also enough differences. In Fig. 10-4B, the same ETVG gestalt function symbols are depicted as in A. Here I have surrounded the group of the symbols of those functions that I assume to belong to one of the four systems described in the FACADE theory:

There are structural similarities: the BCS covers the positive, and the FCS the negative functions of the ETVG gestalt factors Gml, Ll, and Fl. The MP system comprises the factors below Gml, as MP has been attributed to the lateral geniculate nucleus (p.180), to which the gestalt factors Dm and Dl correspond (see Table 3-1). The ORS above BCS and FCS is in fact not dealt with in the ETVG, but is thought to be positioned at the mental evolutionary level above the psychic gestalt factor hierarchy,  

ETVG: Agreement insofar as all that happens in visual perception, primarily happens in the functional sphere. Only when the functions are activated superthreshold, a phenomenon occurs. The ETVG, however, does not make any assumption on the lowest level at which a conscious percept can occur; it but states: every percept, once it occurs, contains all qualities from the lowest level to the highest level being actualized.  

b) Statements about the Boundary Contour System (BCS)

ETVG: O.k. A spatial completion process occurs at least at the fourth level by XLl-laws, particularly the ETVG gestalt law Llc Llc that establishes the "sum of Dls" by the informative effect, and maximizes it by the formative, here the "overshooting", effect of the gestalt function Ll+. A spatial completion process is formatively established also at the fifth level by XFl-laws, which can fully "close" partially closed contours and fields (in agreement with the "law of closure" of the Berlin gestaltists).

ETVG: Yes, as shown in Part 9, the same gestalt factors constitute the MPCs as well as the NPCs. However, the ETVG predicts "real" and "illusory" fields as well.

This corresponds to the findings of von der Heydt, Peterhans and Baumgartner (1984).

ETVG: The development of short subjective contours, perpen- dicular at the end of lines, is described and accounted for particularly by the gestalt law Gmlc Llc in Part 9. Connections across small gaps between these short contours to achieve a longer subjective contour by the gestalt law Llc Llc, are predicted as well.  At the fifth level, the closedness factor can contribute to the increase in the subjective contours' length by closing the contour.

c) Statements about the Feature Contour System (FCS) and its relationship to the BCS.

ETVG: One brightness (color) spreads across the one area, the other brightness (color) spreads across the other area, each starting from the common boundary between the areas and going "outwards". This is, at the same time, a homogenizing process that tends to "average" all brightnesses and colors in each field, i.e. tends to offset all brightness and color differences. In a figure/outfield system, this process goes on either "inwards" between boundaries in the infield, or from a boundary "outwards" in the figure's outfield. Not a diffusion process is the cause of the filling-in, but a "command from above" (Part 3).

FACADE: Regarding the processes of both BCS and FCS:
        "Either of these processes, by itself, is insufficient to generate a final percept." (p.174)
        "The rules governing either process can only be discovered by studying how the two processes interact" (p.174)

ETVG: This is similar to the ETVG view: neither the positive nor the negative function of the same gestalt factor can, by itself, form a phenomenon, which is understandable only on the grounds of the functions' interrelationships (i.e. their antagonism).

 

2. Criticisms of the Grossberg theory

The two readers of both the ETVG and the Grossberg theory seem to be right with their impression of similarity. As just shown, there are indeed a number of similarities, and I believe that the Grossberg theory is, among all theories of sight, that with the most similarities to the ETVG. This is all the more important as both theories were developed independently from each other. In the following, the Grossberg theory will be critically analyzed. Since, however, a theory cannot be criticized on the grounds of another theory, criticisms of the Grossberg theory concern its ability to account for essential facts, its intrinsic logic and its stringent arguments, as it is predominantly these properties that distinguish a good theory.

a) Invalid distinction between "visible" and "recognizable" percepts

A particular distinction is basic to the Grossberg theory: that between "visible" and "recognizable" boundaries:

Grossberg, however, stated that in the abutting grating illusion such as in Fig. 9-9B+C:

This means not only that, according to the Grossberg theory, a solid line between two equal brightnesses or colors is "invisible", but also that the same solid line becomes "visible" as soon as one of the brightnesses/colors on either side of the line is varied. Such a basic distinction is inacceptable, for several reasons:

(1) The Grossberg theory designates the illusory grating line to be a "recognizable" percept, although it is, according to its own definitions, a percept that is composed of both a number of "recognizable" (here: "illusory") and a number of "visible" (here: "real") lines, as the black bars have a certain width and thus a "surface", which establishes at their ends a brightness difference to the adjacent white of the paper. The "surface" property of such lines is assumed by Grossberg himself on the same page (p.619, regarding the "outline square" in his Figure 2A).

(2) The theoretical composedness of a phenomenal line of fundamentally different perceptual qualities contradicts phenomenal fact, as either of the "illusory" contours mentioned above is a phenomenal whole, and thus does not appear "composed". If "subjective" and "real" contours/fields are considered as being only differentiations of a "phenomenal" contour/field, as the ETVG does (Part 9), a non-composed, holistic phenomenon results.

(3) A striking objection to the assumed fundamental distinction between "visible" and "recognizable" percept is demonstrated by Fig.10-5. Nobody would deny the thick ring in Figure A to have a "surface" and thus be "visible" (sensu Grossberg). B and C, too, must be "visible" as they have surfaces as well, although smaller ones. But what about the thin line in D (which could be even thinner)? Here only a "circlelike line" can be seen which has no surface itself, but is a "boundary" between two fields of equal brightness; it is thus "invisible", according to the Grossberg theory. The author does not determine for such series the theoretically relevant stimulus condition that decides on which percept is visible and which is not. There is in fact no possibility for such determination, as there is a smooth transition from the perception of a figure ("ring") to the perception of a boundary of a figure ("disk").

 

Figure 10-5.  Smooth transition from "visible" to "invisible" (only "recognizable") boundaries.

 

Figure 10-6. Different kinds of boundaries

(4) There is, indeed, a difference between at least three kinds of borderlines, as shown in Fig. 10-6: (A) a separate solid, or (B) dashed, borderline (a black one, for example) between two adjacent fields of equal brightness/color, and (C) a borderline which is constituted only of two adjacent fields of different brightness/color. It is justified to designate and account for the three kinds of borderlines differently. However, Grossberg has not done this. According to Grossberg, the boundaries in A and B are considered only "recognizable", i.e. "non-visible", but that in C "visible".

(5) Also possible is a combination of the boundaries such as of Figs 10-5C and 6C (when grey instead of black) that is not mentioned by Grossberg: the infield of a disk with a contour of any color (black) is filled-in with a different color (gray), as shown in Fig.10-6D. We find these kinds of figures in negative afterimages (Figs 6-15,18, 20) and in the case of neon color spread. According to the Grossberg theory, D must, on the one hand, be a "visible percept", as two different colors are separated by a boundary. But, on the other hand, the boundary is not a "genuine" or "simple" border between two different brightnesses, which is proposed by Grossberg to be requisite for a "visible" percept.  Fig. 10-6D seems to contradict the stated sharp theoretical distinction between "visible" and "invisible". Another example for the ambiguity in determining a line to be "visible" or "invisible" is the nature of the illusory Ehrenstein contour; primarily, it is formed by the connected lines perpendicular to the black lines, according to both the Grossberg theory and the ETVG. The enhancement of brightness inside the circular contour is secondary, contrastively caused by the black lines. But is it justified to consider the illusory Ehrenstein contour "visible" only because of this secondary effect?

(6) In the case of Fig. 10-6C, the (sharp) borderline at each location is characterized by a single abrupt transition from one brightness/color to another. In the extra borderline in A, two such abrupt transitions can be seen: one from the (black) borderline to the one field, and the other from the borderline to the other field of equal brightness. One recognizes: not even the cases depicted in Figs 10-5D and 10-6A are phenomenally unequivocal; they are ambiguous to a certain degree insofar as one can perceive the black boundaries as either the widthless boundary of a figure, a "disk", or as a figure itself of a certain width, a "ring", though an extremely thin "ring".

In summary, any distinction between "visibility" and "invisibility" in the above mentioned cases seems unfounded and artificial.

b) Restricted domain of filling-in

The Grossberg theory wants to account for the perception of brightness, color, or inhomogeneities such as lines at locations of the visual field where there are no adequate stimuli, such as in the cases of the "blind spot" and scotomas.

There are a number of objections to the  proposed hypotheses:

(1) The astonishment about extended bright regions in a percept although corresponding stimuli are missing is based on the belief that only the illumination of a coherent group of receptors with equal light energy leads to an extended percept of equal brightness. This belief is wrong, as already the ganglion cells do not (or only very weakly) respond when the center and the surrounding area of their small receptive fields are equally illuminated. It is thus impossible that receptor information simply "flows" through a number of processing stages in order to exhibit extended brightness of a percept. We must therefore already recognize extended functional scotomas at the retinal level. Despite this lack of ganglion cell activation, we are able to perceive the extended bright area of the moon, for example. The Grossberg theory does not propose any hypotheses to account for this fact, but the ETVG does so: these "tacit" ganglion cells are activated by a "command from above", i.e. from LGN cells which were in turn activated by the brain area V1 and so forth. (The complex activation process is described in greater detail in Part 3.) It follows, that the "filling-in" process already begins in the LGN, and not only in the cortex when the "feature contour" system (see Figure 10-4) is exhibited, as the Grossberg theory assumes. (Note that the ETVG does not really speak of "ganglion cell" activation, but of "P" activation (and so forth), that is triggered by a "command from above".)

(2) According to the Grossberg theory, filling-in is a diffusive process:

One cannot expect this procedure of activation via diffusion to have already been finished prior to the very short space of time in which the percept (such as the "moon in the sky") needs to be established. (Köhler allowed several minutes for the establishment of a syncytial information flow.) The activation "upwards and downwards" through a short series of hierarchical stages, as proposed by the ETVG, is a much faster procedure.

(3) There is also a counterprocess to "filling-in", which can be observed in the actual-genetic processes of negative afterimages. An example is when the dedifferentiation process (shown in Fig. 6-18 D1,2,3) reverses, and the gray infield of the disk fades, i.e. its blackness "retires" to concentrate itself in a ring area. See also intermediate stages of partial filling-in in Fig. 10-6D as well as in Figs 6-18 and 6-20.

(4) These filling-in processes and their counterprocesses also occur under "weak" stimulus conditions in original sensory perception, such as in the "actual-genetic" and "actual-lytic" processes demonstrated in Figs 5 to 16 in Part 6.

(5) Under certain conditions, one can observe the blurring of contours in filling-in processes and the sharpening of contours in "emptying" processes. Grossberg, however, dealt only with "sharp" contours and did not explain what a filling-in process looks like in the case of "blurred" contours. As seen in negative afterimages, an inner contour can smoothly disappear by blurring, i.e. by losing contour property and becoming field property. Also the reverse process often occurs. This is a further reason against the assumption of two "distinct systems", one processing the borderline, and the other processing the fields, even if they are considered to be "intimately connected". The ETVG suggests the assumption of a single process that creates both the borderline and its adjacent fields, which allows line and field to undergo mutual transition to each other. Phenomenally, line and field are polarly related to each other, as shown in Part 2. The Grossberg theory also does not seem to say anything about either why the blurred projection of sharp object edges leads to sharp figure contours, or why the percept lacks distortions despite distortions in the object's retinal image. The Berlin gestaltists can account for these facts by means of their "internal forces", whereas the ETVG proposes the "formative effect" of the gestalt factors to account for them.

c) The lack of a two-dimensional figure/outfield concept.

The purpose of a neurophysiological theory of sight is to show the neurophysiological conditions of visual phenomena. It especially should account for object perception, which is the foundation of object recognition. Our visual percepts represent the three-dimensional human environment, and are thus three-dimensionally structured. But an "object in its surroundings" is projected onto either retina as a two-dimensional image whose two-dimensional perceptual representation is the "figure in its outfield". If one wants to account for three-dimensional experiences, one must first account for the two-dimensional experiences, or at least account for the processes assumed to be the condition of these experiences. Two-dimensional figure/outfield percepts are not only theoretically required; they are, after all, well known facts since Rubin (1921), and have belonged to the empirical basis of perceptual psychology since the twenties. If these fundamental phenomenal facts are ignored, one cannot develop an acceptable theory of sight.

Grossberg deals with two- and three-dimensional percepts, but he does not propose any concept for the basic two-dimensional figure/ outfield phenomenon, i.e. he never describes, or even defines, its properties, parts, or aspects. He does not show the basic interrelationships within the figure/outfield system, particularly

  1. that a figure is composed of a contour (borderline, "boundary") and a field that is enclosed by the contour, which can be designated infield;
  2. that the figure is in turn enclosed by another field, the outfield;
  3. that the contour "belongs" to the infield and not to the outfield;
  4. that every contour (borderline) borders off two fields from each other, but does not necessarily enclose one of the fields; in the case of non-enclosedness, both fields have equal "weight", whereas in the closure relationships they have unequal "weight" and properties.

Grossberg stresses the borderline, even with two terms: "boundary" and "contour", but he does not propose any concept, or at least a particular term, for the things that are bordered off with the boundary. Such a thing would be best designated "field" or "area", as both terms include a particular plane in these definitions, which is the most crucial property of a "depthless thing". To prevent a linguistic misunderstanding, or a "theorist error", I quote the respective definitions from "Merriam Webster's Collegiate Dictionary" (1995), which show the main properties of "area" and "field" as meant in the present issue:

"Area" = "the surface included within a set of lines" - "a particular extent of space or surface or one serving a special function" - "a geographic region".

"Field" = "a region or space in which a given effect (...) exists" - "a space on which something is drawn or projected" - "the area visible through the lens of an optical instrument".

As just shown, "field" and "area" have a common semantic field; they are thus best suited to express a plane bounded by borderlines and perhaps even having a particular function. These terms are suitable also because they are familiar to visual scientists, especially neurophysiologists, from the terms "cerebral cortex area", "receptive field", and "field of vision". Yet precisely these two words, most suited as technical terms in a theory of sight, are not used in the Grossberg theory to designate the fields bordered off by a boundary. Searching for "area" and "field" in the more than 40 pages (pp.116-125, 141-148, 173-184, 471-474, 627-629, 2659-2665), on which the Grossberg theory is described, I found "area" in the sense of cortex area and of a part of a diagram, and I found "field" used in the terms "receptive field", "gated dipol field", and in "multiple cellular fields"; once, however, also as "color field", the only example on these pages with the meaning I searched for.

Grossberg does not inform us why he avoids precisely these two most significant words. He also does not provide any clear, theoretically relevant conceptual information on the infield-contour- outfield system, particularly on the relationships between

because his theory does not include any concepts regarding (two-dimensional) "figure", "infield", and "outfield".

It puzzled me that Grossberg would propose such an unclear theory. The theory does agree with a number of neurobiological facts, especially as it starts from these facts. Empirical investigations suggest that there are two relatively distinct and parallel parvocellular processing streams, of which one processes the boundary, the other brightness and color, and both are interconnected. But there is another group of facts, the visual phenomena that are predominantly organized as two-dimensional infield-contour-outfield relationships. Grossberg must have realized that the neurophysiological facts are not sufficient to account for these basic figure/outfield facts, as no neurons have yet been found that detect "fields". Instead of calling for research to find such neurons, he developed a visual theory on the grounds of contemporary, and thus naturally still incomplete, neuro- biological results. Obviously, such a theory could never become an acceptable one, and so he may have unknowingly obscured those phenomenal facts he could not account for with already known cells, especially the structure of the two-dimensional infield-contour-outfield percept (which is the foundation of a three-dimensional figure/ground percept).  

If one wants to develop a theory whose untenable nature cannot be recognized at first or second glance, one is easily led to unconsciously take special precautions. Grossberg did not even incorporate a "field" concept into his theory (apart from not establishing an "infield" and an "outfield"). Fields, however, are fact and are necessary to account for the changes in brightness and color that can be observed on them, such as in the "filling-in" process. He was therefore compelled to neglect both the real existence of phenomenal fields and their relationships to their boundaries, although these facts must be considered in every theory of sight. He replaces the terms "area" and "field" with a number of vague and nebulous words that have a wider meaning, and uses them as well when he feels he cannot escape indicating the "things" (figure, infield, or outfield) that are actually filled-in with brightness or color. Thus, we encounter words such as "region", "domain", "spatial domain" (p.146, 182), "sides of the contour" (p.618), "perceptual domain" (p.173), and "perceptual space" (p.186), all more or less far from suggesting that we are dealing with the topic of visual perception at all and especially with that part of the visual field that is bordered off from an enclosing field by a set of lines. We also encounter other expressions, such as "interior of a patch" (p.144), "inside the Ehrenstein figure", instead of something like "infield" which clearly designates a border-enclosed figure field. Even statements in which "closure" is suggested are kept vague:

First, in choosing to use the plural "regions", the author introduces vagueness (which even the addition of "connected" cannot remove); second, once more he does not propose that the boundary is responsible for separating figure and ground, but the "filling-in" (which often is identical to brightness and color).

Grossberg uses peculiar names for the two processing systems he assumes. No "field" appears, but a term for the opposite of a field, the borderline ("boundary" and "contour"), is used three times; the relevant term "field" is replaced by the irrelevant term "feature". Simple and distinguishing designations such as "Boundary System" and "Feature System" would be sufficient, but are too "dangerous" to be used, as perceptual psychologists could too easily suspect that "feature" is not the right counterpart of "boundary" in a visual figure/outfield percept. This danger can be circumvented by blurring the distinction, such as by adding a third word to both system names, most effectively one meaning the opposite of "field". Thus, "Boundary Contour System" and "Feature Contour System" come into being. Confusion is complete. Functional descriptions of these systems are often given, but do not decrease the confusion, as they are confusing in themselves. I might not be the only one to have had difficulties in understanding the relationships between BCS and FCS, as the attempt to dispel misconceptions (p.146) suggests, an attempt that does not seem to me to have been successful.

This (surely unconscious) creation of general confusion serves its purpose: one is fully occupied with the desire to understand the relationships, and thus suppresses all other thoughts, such as those of "fields". Further, smaller opportunities are given to escape field-awareness, as when "filling-in" is simply spoken of without indicating what exactly is filled with brightness or color. Or when "border-ownership" (p. 621) is mentioned in respect to a line without naming the owner itself, which would here be the boundary of a two-dimensional figure. Not only "contour" is non-distinguishing in the systems' names, "feature" as well, as "feature detectors" are involved also in the boundary contour system (p.146).  

In his second theory, Grossberg goes two steps further, i.e. farther away from two-dimensional infield-contour-outfield perception. The first step comes when he coconsiders three-dimensional percepts and elevates "figure-ground" to a key concept, as is indicated in the titles of a number of articles (Grossberg and Wyse 1991, Grossberg 1993, 1994, 1997, Grossberg and Pessoa 1998). As he directs the readers' attention toward three-dimensional perception (an advantage of which is that at the same time it draws their attention away from any two-dimensional figure/outfield concept), he is given an additional opportunity to use further terminology that prevents the reader from thinking of two-dimensionally extended fields or areas: "surface", "surface regions" (p.618), and more often "surface representation". However, "surface" (and even "flat surface", p.618) is not the same, as a "field" (infield and outfield) is extended on a frontal plane, but the perception of a "surface" can only be achieved by means of a view perpendicular to the frontal plane. The second step involves "form". Form is the differentiation of a figure; a (two- or three-dimensional) "figure" concept is the reference system for any "form" concept. Grossberg cannot propose a form concept built on a figure concept, since he does not propose any figure concept. Grossberg designates his second theory, with which he wants to add depth and form to his theory, "FACADE" Theory; FACADE meaning "Form-And-Color- And-DEpth" (p.471).  

Introducing "depth" into the theory multiplies the problem of defining what is really meant by the terms used. The difference between "field" and "surface" on one hand, and "features" (brightness and color), on the other, has already been mentioned. The solution of the "field"-problem in the case of "feature" is threefold: first, the referent of the concept "feature" has expanded beyond brightness and color: "the filled-in representations combine properties of surface depth, position, orientation, brightness, and color" (p.466); second, "feature" is less often used from this point on, and is replaced by "properties", "surface properties" (p.628, 2659), and even "continuous surface properties" (p.644); third, the previously extensive use of the term "feature" is avoided simply by concealing it in the abbreviation "FCS". However, this brings another "danger" with it: the consideration of three-dimensional objects themselves requires a comparable consideration of several, at least two, two-dimensional figure/outfield percepts, extended in different depth planes. For this, a new abbreviation appears: "FIDO", which means "Filling-In DOmain" (pp.475, 628, 2663). From now on "Monocular FIDO" is distinguished from "Binocular FIDO".

It is understandable that a theorist - I know this from the development of my own theory - in the beginning tries to collect all facts that confirm his hypotheses. But the more facts agree with the theory, the less attention may be devoted to contradicting facts. Many theorists fall victim to this danger, among them the early gestaltists including Köhler and Koffka, and, I have to assume, Grossberg as well.

back to contents
home