Chapter 8 Chapter 6 The Moment of Perception: A Theory of Vision

"Psychology is a very unsatisfactory science." —Wolfgong Kohler The decay time of icon memory and working memory can be quite short.What do we know about the timing of the various processes leading to consciousness?Recalling the content of the second chapter, we know that some cognitive scientists like to regard the activities of the brain as the process of performing calculations. They believe that what causes consciousness is not the calculation itself but the result of calculation. Some have claimed that certain brain activities do not reach a level of consciousness unless they last longer than a certain minimum time.If the activity is weak, this time may be as long as half a second.Simply to guide our quest for the neural correlates of consciousness requires an understanding of the type of duration of brain activity that corresponds to a single "moment of perception."What type of time is involved in a single processing cycle?

Let us consider the following situation.First, subjects were presented with a 20 ms long transient red light stimulus.Immediately thereafter, a 20-ms green light stimulus was presented at the original place.What did the subjects report seeing?Instead of a red flash followed by a green flash, what he saw was a yellow flash.Just like what you see when the two colors flash at the same time.However, if the green flash was not immediately followed by the red light, the subjects reported seeing the red flash.This suggests that the subjects were unlikely to be aware of the yellow color until the information from the green light had been processed.

Therefore, you cannot feel the real onset of a stimulus, nor can you estimate the real duration of a brief stimulus.As early as 1887, the French scientist A. Charpentier discovered that a flash stimulus of 66 milliseconds does not seem to last longer than a 7 millisecond flash stimulus. In 1967 the American psychologist Robert Efron wrote a good and insightful article on this issue.By estimating it in different ways, he concluded that the duration of the processing cycle is about 60 to 70 milliseconds.This number is for salient stimuli that are easier to observe.It is not surprising that the processing period will be longer for unclear or more complex stimuli. .

So, how much time is needed for more complex processing?In such cases, presentation of a visual stimulus is usually followed by a quick mask, that is, presentation of a visual pattern at the same location in the visual field, to interfere with some of the processing necessary to view the original stimulus , explaining this result in detail is difficult.If the system is simple and sequential, with signals progressing steadily from one stage to another without pauses, and taking no time to enter consciousness, then it is impossible for the signal from the mask to catch up with the signal from the stimulus.Since masking can interfere with the perception of stimuli, this implies that at least some processing steps take time.This is possible anyway.Despite the difficulties of interpretation, masking effects can provide us with some useful information about this process.

The American psychologist Robert Reynods investigated this question through several experiments.He wishes to show that different aspects of perception can be seen at different times.In other words, he sought to study the time course from stimulus presentation to relatively stable perception. As an example, let us look at the time of formation of the perception of phantom outlines described in Chapter 4.To prevent the subjects from guessing or lying, Reynolds presented the subjects with one of the two patterns in Figure 22.Each pattern is composed of three notched discs as shown, where the first illusion border is straight and the second is curved.Stimuli were presented for 50 milliseconds, followed by a mask as shown in Fig. 22c after a certain delay ① time.The stimulus pattern is large and bright, and the subjects can clearly see the three notched discs even if the presentation time is very short. Due to the existence of icon memory, in the absence of masking, we have reason to believe that the signal from the displayed image has a positive effect on the brain. The action time will exceed the graphic flashing time by 50 milliseconds (probably hundreds of milliseconds).

Reynolds found that the hallucinatory triangle was invisible to the vast majority of subjects if masking followed the stimulus.The few people who report seeing hallucinatory triangles also often make the mistake of confusing straight and curved triangles.However, with a latency of 50 to 75 milliseconds, or an SOA of 100 to 125 milliseconds, all observers reported seeing triangles, although they could not yet tell with complete accuracy whether the sides of the triangle were straight or curved. This clearly shows that the total processing time depends entirely on what he sees.Three notched disks (pacmen) can be seen clearly for a period of time before the phantom triangle appears.

It should be noted that these experiments cannot precisely explain when the brain produces the "neural correlates" of perception.It can only mean that some aspects of perception are likely to take longer to process than others. Reynolds conducted another, more complicated, similar experiment.The same illusionary triangle is drawn as if placed behind a transparent brick wall.The interpretation of such a visual pattern is uncertain.The subjects first saw three notched discs, followed by a bright triangle, which was rejected again, and then triangle perception reappeared①.Each of these last three phases takes about 150 milliseconds.

Obviously, the timing of "calculations" depends on their complexity.Although a detailed explanation still depends on understanding exactly how signals are transmitted between different brain areas and how they interact (which is unlikely to be simple), at least for now we have a good understanding of the various types of timing required for visual processing. Got a rough idea.Until we have a clearer understanding of the different brain processes involved in seeing and how they interact, it is unlikely that we will get a more precise time. I have briefly described many aspects of visual processing, but have not systematically explained how we should recognize all these processes, which is a difficult problem.If this were a book devoted to visual perception, I would have to devote some space to describing some of the latest thinking about vision—that is, how the brain performs complex activities that enable us to see the outside world.With the exception of the cognitive scientists mentioned in Chapter 2, most theorists have shown little interest in consciousness, and for this reason, coupled with the fact that there is not yet a generally accepted theory of vision, many different approaches to No detailed description was given.However, the following short overview will give readers a general impression. ①

People are interested in vision for many different reasons.Those who wish to create a visual machine that can see as well as we do or better than we do, in order to apply it to domestic, industrial, or military purposes, do not much Concerned about how the brain does this job.A visual machine doesn't need to closely mimic the human brain any more than an airplane needs to flap its wings. Others are primarily interested in how humans see objects.Some functionalists take an extreme view.They argue that knowing the details of the brain will never yield anything useful (2).The idea is so outlandish that most scientists wonder why it even exists.The other extreme point of view comes from some neuroscientists, who are mainly concerned with the response of animal brain nerve cells to visual images, but pay little attention to how this activity produces vision.Fortunately, there are now a small number of students of vision whose views lie between these two extremes, who are interested in both the psychology of vision and the behavior of nerve cells.

People's ideas on these questions also vary widely, and some people think that it is important to study the visual environment-that is, the earth under our feet, the blue sky above us, and everything in between.They don't care about the brain because they think that all that needs to be done is to "resonate" with aspects of the environment, whatever that means.They refer to themselves as Gibsonians.It is named after its late master Gibson (J.J.Gibson).Others have attempted to analyze basic, but rather limited, operations of vision, such as shape restoration from shadows, the illusion of a barbershop sign, etc., and to write computer programs that solve these problems.In the field of artificial intelligence, this tradition is still strong.Still others liken processes in the brain to objects or events in everyday life.They often talk about things like "searchlights" or "opening a file for something", and over the past two or three decades, the explanations used have often been based on how computers work.He uses a set of well-defined rules to obtain the desired conclusions, and involves certain computer concepts, including central processing, random The interactions are roughly parallel, and there are no clear rules. (Discussed more fully in Chapter 13.)

As we saw in Chapter 4, Gestalt psychologists wish to uncover the fundamentals of visual activity.They argue that just as understanding the laws of aerodynamics is important to understanding the flight of birds and airplanes, understanding vision must look to the universal principles it involves.Modern forms of this research approach often express their theories in informatics terms.Not surprisingly, mathematicians tend to discover some general mathematical principles.To the average reader, describing all these ideas would probably require a large book. All of these points of view have some value, but they have not yet been brought together to form a detailed, widely accepted theory of vision, so long as the problem of visual awareness is sidestepped.Any existing theory of vision is inadequate, and in any case vision is a complex and difficult process, and it is unlikely that we will come up with a comprehensive theory of vision until the next century.If we want to study the problem of visual consciousness now, we have to do our best.For this, we need some tentative perspective, or we'll just miss the boat. I think the research methods developed by the late Devid MaIT, a young English man who took a degree in mathematics at Cambridge University in preparation for brain research, are very useful.His doctoral dissertation presented a detailed and novel theory of the cerebellum.Later, sydney Brenner and I provided him with an office in our laboratory in Cambridge, UK, where he developed a general operating theory of the visual cortex and hippocampus, part of his interest Turned to visual artificial intelligence, and went to the Massachusetts Institute of Technology (MIT) to cooperate with the Italian theorist Tomaso Poggio (Tomaso Poggio), in April 1979, the two of them went to the Salk Institute (Salk lnstitute) A one-month visit to me.Marr once wrote a book called Vision (published posthumously).In the book, he explains many innovative ideas about vision in a concise way (his scientific papers are not easy to read).While not all of these ideas have stood the test of time, the book's treatment of these issues remains ingenious for its time.The final chapter includes a hypothetical conversation between Mal and a reluctant believer (myself) that loosely mimics the many conversations the three of us had between him and Poggio when they were at Salk. Marr envisioned a general framework for describing a rough outline of the visual process.He believes that the main task of vision is to obtain the representation of shape; lightness, color, texture, etc. manifested in many ways.Marr argued (and, of course, was largely correct) that all this could not be done in one step, and instead postulated the existence of a sequence of representations.He called them "raw feature map", "2.5D map" and "3D model" representations. Primal sketches reveal important information such as light intensity changes, geometric distributions, and organizations in 2D images.The features it handles include boundary line segments, blobs, endpoints, discontinuities, and boundaries. The 2.5-dimensional feature map visualizes the orientation (and approximate depth) of visible surfaces and their contours in an observer-centered coordinate system. The 3D model representation describes various shapes centered on the object and its spatial organization. In this way the vision task can be divided into at least three independent stages.This is very rewarding because it at least makes us realize that there is so much more to seeing.But it can't all be right in the details.The three stages may be only a first approximation, eg color, texture, motion should be added on top of "shape".There may be more than three stages, and these stages of processing may not be as strictly distinct as he describes, and there may be two-way interactions between them. However, his framework illustrates what happens when we look at objects. Processing type, (I discuss its relation to neuroscience in Chapter 17.) Marr's untimely death of leukemia at the age of 35 was a major loss for theoretical neurobiology research.I firmly believe that if he is still alive, he will never rest on his laurels, but will further develop his brain theory with the progress of research, and his ingenuity and imaginative creativity will definitely help us break through all the difficulties we face today.Because he not only has extraordinary intelligence, but also has a strong ability to digest and absorb a large amount of experimental evidence in different fields. What explanatory style do we need in order to understand the brain?My own view is closest to Rama's utilitarian theory of perception, which holds that visual perception involves neither the kind of rigorous, intellectual reasoning that we argue with, nor the brain's response to visual input." Resonance" kind of vague idea.Visual perception also doesn't require solving complex equations to solve, as AI researchers often imply.In contrast, he argues that perception "uses rough rules of thumb, shortcuts, and tricks of certain dexterity. These have been acquired by trial and error through eons of natural selection. It is a familiar strategies, but for some reason escaped the attention of psychologists who seem to have forgotten that the brain itself is a biological organ...." I also agree with Rama Tsanjun's statement: "Just open the black box to study the response of nerve cells is the best way to address this problem. But psychologists and computer scientists are often skeptical." According to Rama Zanjun, the main task of visual psychologists at this stage is not to construct complex mathematical theories to explain their results, but to outline the so-called "natural history" of vision, especially the ten-level stages of vision.When a visual task is broken down into its many components, especially when certain interactions are shown to be weak or absent, we know exactly what needs to be explained in neuronal terms.These explanations do not necessarily involve complex mathematical theories, but must involve the properties of the interacting neurons and the details of their interconnection.Therefore, due to the complexity of the visual world, one expects to find rough but efficient fast processing processes with multiple dynamic interactions. The next step is to understand the human brain (and the monkey brain) and the many nerve cells and molecules that make it up, which will be the subject of Part Two. ① Reynolds reported his results using the term "stimulus onsel asynchrony" (SOA).Since the duration of the stimulus is 50 ms, an SOA of 50 ms means that masking begins immediately after the end of the stimulus.I call it zero latency. ① Note that the subjects did not report all these phases in one experiment.The present results are inferred by comparing perceptions after different masking delays. ②I relegate some of Libet's research work to Chapter 15 for consideration. ①Of course, for those who conduct visual awareness experiments, the most important thing is to have a detailed knowledge of visual psychology and various visual perception theories.In this way, at least undue mistakes can be avoided. (2) "All you need to know about the brain is how to simulate it." Philosophers, AI experts, and linguists often take this view.This view is not new among those who eschew the rigorous scientific method.