The first of this pair of manuscripts offers an approach to modeling illusory contour formation that incorporates many of the aspects of neural models but also includes feedback-- higher levels of processing can influence the lower levels of processing.
This manuscript suffers from several serious problems. The central problem is that, overall, there is too much unnecessary detail. It would make a good introductory chapter for a book on this approach, but it does not offer the readers of this journal a complete model or a novel approach to modeling. The author acknowledges that many other models use similar principles, yet I found it hard to tell what predictions this model would make that were different from other models.
Other problems include:
A discussion of the philosophical benefits of this approach that is problematic (the author suggests that this approach offers a way around concerns raised by Dennett and ORegan yet goes on to assume that there is a subjective percept -- a concept that both Dennett and ORegan argue against in favor of an interactive, temporally evolving, process that allows appropriate actions to occur; The author may want to consider a recent target article in Behavior and Brain Sciences by Pessoa, Thompson, and Noe that discusses these issues).
There are numerous assertions that are not based on data (e.g., page 2: ...figure reveals ...processing in which global features are detected as a whole.... How do we know that the individual inducing elements are indistinguishable from the camouflage elements? What about pairs of elements? Would local edge alignments constitute local features or global features? The author may want to consider a series of experiments by Rock and Anson (1979, Perception and discussed in Rocks book The logic of perception) on hiding illusory contours (for example, making the inducing elements similar to other adjacent elements). On page 5: the appearance of an illusory figure despite an overall brightness difference between the inside and outside of the figure does not necessarily mean the edge is amodal. The edge may be seen because of local differences in brightness or texture density. The author may want to consider work by Kellman and Loukides (reported in the 1987 Petry and Meyer book Perception of Illusory Contours) demonstrating that global brightness differences are not necessary for illusory contour formation, and Kellman and Shipley (1991, Cognition) suggesting that amodal and modal completion are related).
A critique of Grossbergs approach, but a focus on early papers and not more recent work (e.g., FACADE theory).
Finally, the account offered for simultaneous contrast may be consistent with some accounts, but is not universally accepted (the author may want to consult recent work by Allan Gilchrist on this issue). minor concerns Assertions that the subjective percept is richer in information or has more explicit spatial information does not conform to the familiar usage of the term information in perception.
The second manuscript describes in detail a model of perceptual completion that allows for long range effects using feedback and repeated iterations of edge finding and spreading algorithms. The attempt to provide a new account of illusory contours that borrows from existing neural models and is constructed with perceptually meaningful representations is laudable. In evaluating any model one must consider how well it accounts for existing research and whether or not it makes predictions -does it lead to novel lines of research. Neural models can account for a number of findings, but it is quite hard to tell if these accounts require specific parameters. Is it possible to account for many phenomena with a single set of parameters? When faced with complex models, such as neural-based models and the feedback-based model presented here, it would be useful to understand what predictions are independent of the particulars of the parameters--are there differences between this --> --model and Neural models that are parameter independent? It would be useful to present simulations side by side to illustrate any qualitative differences (as is done in Figure 4). Additionally, I am not sure that I can see novel predictions made by this model. Note, parsimony is not a valid metric for comparing the two types of models since the nervous system need not follow any rule of parsimony.
The author cites a few basic findings from the illusory contour literature on how the perceived contour changes with select spatial parameters. Although the models qualitative behavior mirrors the psychophysical data, it is not clear that it will do so quantitatively. For example, the author cites Shipley & Kellman and Banton & Levis finding that illusory contours are a function of inducing element length and illusory contour length, yet it is not clear that this approach predicts the specific trade off found between inducer element length and illusory contour length that is referred to as support ratio. There is some more recent work on this issue by Lesher and Mingolla (1993, Vision Research). Additionally, the model predicts contrast will affect illusory contour strength, but is there any reason that overall illumination should also affect illusory contours (Bradley and Dumais (1976, Perception and Psychophysics), found that strength was inversely related to illumination levels). For an earlier discussion of other parameters that influence illusory contours see Halpern et al (1983, Perception). In addition to parametric effects, a good model of illusory contours should account for cases where contours are not seen. Does this model account for cases where local edges are aligned yet illusory contours are not seen by humans? What would this model do with: line drawn inducing elements and four cross shaped inducing elements placed at the corners of a square?
minor points: 1. On page 13 it is asserted that illusory contours fade toward the center of the edge. I am not sure I have this experience, please support this claim with a reference. 2. The author should consider discussing recent work by Field on grouping.
In both manuscripts the small number of citations and lack of references to related empirical research suggest that the development of this model was constrained a few phenomena. In the early stages of theory development it is sometimes appropriate to set aside a few phenomena that cant be explained. However, given the amount of research on illusory contours it would be appropriate to try to incorporate as much as possible before presenting a new model.
I would recommend that both Computational Implications I and II be rejected and that the author be encouraged to resubmit a manuscript that incorporates a brief discussion of the general issues raised in Computational Implications I along with many of the details presented in Computational Implications II. This new manuscript should be shorter than either of these two, and it should include clear comparisons between the new model and existing models.