Chapter 13 Chapter 11 The Primate Visual Cortex

"We should keep things as simple as possible, but not too much." --Albert Einstein The primate cerebral cortex consists of left and right sheets, and each sheet can be divided into many different cortical regions.How to determine whether a specific area on the cortex belongs to the same cortical area?There are many different criteria that may be valid.The first is to look at the shape of the structure in cross-section under a microscope—for example, whether it has an extended fourth layer.We have observed streaks that clearly define zone 17.This simple difference is only useful in a few cases, although that changes when more molecular probes are available.Another approach is to find the boundaries of a visual region by examining details of its visual map.But this approach is often less applicable, especially in higher-level visual areas, where most have little retinal counterparts—that is, they do not have simple visual projections.Currently the most effective approach is to look for characteristic patterns of connectivity (both inputs and outputs) for each putative region.The application of modern biochemical methods allows this method to obtain fairly reliable results.But as we saw in Chapter 9, most of these methods are not applicable to the human brain.

Many scientists have contributed to the functional division of the cerebral cortex, especially in cats and macaques.Even so, our knowledge is still incomplete and can only be seen as a preliminary result. Let's start with the striate cortex (area 17), which is now called area V1 (ie, the first visual field). Region V1 is quite large, with nearly 250,000 neurons per square millimeter of subsurface.In the cerebral cortex the number is usually around 100,000, with the exception of area V1.The V1 region on one side of the macaque brain has a total of about 200 million neurons.This is comparable to the approximately millions of axons from the lateral geniculate body.From these numbers we can immediately see that there must be a lot of processing on the input from the lateral knee body to v1. The V1 region is no thicker than the adjacent V2 region, which has a lower surface density.This means that, on average, neurons in V1 are quite small in size.This gives the impression that evolution has crammed as many neurons into V1 as reasonably possible.

Excitatory input from the lateral geniculate body goes primarily to layer 4, with some passing to layer 6 as well.Layer 4 has several sub-areas.Most of the inputs from layers P and M of the lateral geniculate body go to different sublayers of layer 4, and the axons of all inputs are widely branched, so that a single axon may contact thousands of different neurons.Correspondingly, each neuron in layer 4 receives input from many different axons afferent.Nonetheless, only a fraction of synapses (perhaps 20%) in a typical spiny astrocyte receive input directly from the lateral geniculate body.Other synapses receive input from elsewhere, primarily from synapses in other nearby neurons.In this way, layer 4 neurons not only listen to what the lateral geniculate body has to say, but also talk extensively to each other.

Just as input from the retina is mapped to the lateral geniculate body, input from the lateral geniculate body is also mapped to area V1.Of course, this is a mapping of the contralateral view.But this mapping is not uniform (Figure 45).The space corresponding to the vicinity of the center of gaze is much larger than the periphery of the visual field.It reminds me of a humorous map that was popular a few years ago, depicting America through the eyes of a New Yorker.Most of it is the Manhattan area.New Jersey is greatly reduced, while California and Hawaii are only incidentally marked in the distance.

Furthermore, at small scales, the cortical mapping is extremely disorganized.With projections to the cortex through the lateral geniculate body in both eyes except the blind spot and away from the periphery, these two connective pathways to layer 4 separate into random stripes like fingerprints (Fig. 46). ① In the layers above and below layer 4, there are a series of "spots" along the center of the stripe (shown by cytochrome oxidase staining).The neurons here are particularly sensitive to color and brightness. In general, different neurons in the V1 area of ​​the cortex are sensitive to different objects.Recall that neurons projecting from the lateral geniculate body to the cortex have small receptive fields with central-peripheral antagonism, and some neurons in layer 4 in macaques still maintain this property, but the receptive fields are slightly larger.In the 1960s, David Hubel and Torsten Wiesel (both of whom later worked at Harvard Medical School) discovered that for most neurons in layers other than layer 4 in the V1 area For this element, the optimal stimulus is a thin bright (or dark) rod or edge: not a point of light, (For this discovery and other work, Huber and Wiesel won the 1981 Nobel Prize. ) they responded better to the motion stick than to the light and dark flashing stick.For any given neuron, it fires most intensely in response to a wire or rod stimulus with a particular orientation.If the orientation of the stick is only 15 off.Often the firing rate of the cells also becomes very low.Different neurons have different optimal orientations, however, except in some parts of layer 4, neurons immediately adjacent in the direction perpendicular to the cortical surface tend to respond to the same orientation.This is often referred to as a "columnar" arrangement.Furthermore, if one traverses the cortex horizontally, one finds that the optimal orientation changes fairly gradually, with only occasional abrupt changes.In any small area of ​​the cortex about 1 mm in diameter, the receptive fields of all types of neurons often overlap to some extent, and have all possible orientations.This arrangement has been described as "supercolumns" and "cortical modules", but don't take this view too literally.Unfortunately, this formulation is too popular for theorists.Some of them should understand better.

Huber and Wiesel discovered two types of cells towards selection, which they called "simple cells" and "complex cells". The excitatory and inhibitory regions of the receptive field of simple cells are easy to define, and this layout makes it sensitive to A stick or edge responds best.Some receptive field scales are finer than others, and thus reflect finer features. ① The difference between complex cells and simple cells is that their receptive fields cannot be simply divided into excitatory and inhibitory areas.For them to fire, they also need a rod or edge with their optimal orientation within their receptive field, but they are not sensitive to the position of the stimulus within the receptive field.Their receptive fields are usually slightly larger than those of neighboring simple cells, and some complex cells can respond to more complex stimuli, such as a pattern of light spots moving in the same direction.

How do simple or complex cells set up their input connections to produce the observed behavior?It should be soberly aware that after nearly thirty years of research.We still don't know for sure.From the perspective of logic, the problem seems very simple.For simple cells, it fires only if the majority of the set of stimulus points summed up to form the best-response rod is sufficient to produce a response.They perform an AND operation, but require crossing a certain input threshold to cause firing, in contrast, when this or that straight line (which has a similar orientation) is presented somewhere within a complex cell's receptive field, the cell will be issued.It's as if a complex cell takes input from a complete set of similar simple cells and OR's it.It appears that complex cells are indeed processed further than simple cells, but in-depth studies have shown that this simple view leads to difficulties because many complex cells have input directly from the lateral geniculate body.There is also the problem that the best response is usually in a straight line of motion.Sometimes a neuron responds much more to movement in one direction (perpendicular to a line) than in the opposite direction.

It is especially regrettable that this problem has not been resolved.At least one possibility is that simple cells perform an AND operation, followed by an OR operation performed by the complex cell, a common strategy used by all regions of the cerebral cortex.If this is the case, it is very important to understand it. Neurons in the V1 area of ​​the cortex respond in a variety of ways.As we have already seen, many neurons in layer 4 are of the center-periphery type.The same goes for the neurons in the blobs.Most other neurons are orientation-selective, but some respond best to not too long straight lines (often referred to as endpoint inhibition)(1), while others, such as many neurons in layer 6, respond best to Very long straight lines respond best.

Another type of nerve, which receives input from both eyes first, fires strongest only when the input comes from neurons that do not correspond exactly to each other on the retina.This is necessary when extracting distance information for objects in the field of view, because objects at different distances produce different disparities (this was explained in Chapter 4), and we have seen that certain neurons are sensitive to motion in a specific direction, There is no response to movement in the opposite direction.Many of these cells are located in a thin layer called 4B.Many neurons respond equally to all wavelengths of visible light, while others, particularly those in puncta, have selective sensitivity to wavelengths in their central and peripheral receptive fields.In short, they are color sensitive.All of this suggests that different neurons in V1 process incoming visual information in different ways.

The receptive field is the portion of the visual field within which changes in light cause cells to fire.However, the receptive field has a much larger peripheral region in which light changes do not themselves cause cell firing, but mediate the original effect produced by the receptive field.This region, now called the "non-traditional" receptive field, introduces an important insight into the context of the local environment.This environment can have specific characteristics.A cell is not only sensitive to a particular feature, but also to neighboring similar features.This neurobehaviourally important property is likely to emerge at all levels of the visual hierarchy.It may have important psychological implications, as psychologists have found that context is important in many conditions.

Why does cortical V1 have a map of the visual field (although this map is rough and distorted)?It's not because there's a dwarf watching it—our astonishing hypothesis argues against that.The most likely reason is that it keeps the brain's wiring shorter. A neuron in V1 is primarily concerned with what's going on in a small area of ​​the visual field, and it needs to interact with some other neurons to extract the information they express, a rough mapping that keeps them fairly close to each other.Theorists point out that this shortest wiring requirement could also explain various types of chunking found in the cortex, since it allows multiple submaps within an overall main map.A small piece of a submap may have strong interactions internally while having slightly longer connections to neighboring parts within the same submap.Such patches may also have weaker local connections to neighboring parts of other types of submaps.In the same way, it is sometimes useful to think of a city as consisting of many interacting local associations with common interests.How these groups are arranged is partly to facilitate communication, so that there are many supermarkets scattered throughout the city, and every resident is not too far from one of them. Ultimately the economics of this connecting line need to be determined at all levels.Linking this issue to the need to keep the total number of neurons in the neocortex to an appropriate minimum may well explain general laws of cortical (and visual system in particular) organization. The mapping of V1 and other regions is structured in such a way that it appears that its large-scale properties (for example, which area of ​​V1 corresponds to the macula) may be fixed during brain development under the direction of the genes involved of.The specifics of the mapping arise from the modulation of inputs from the eyes, which seem to depend on whether the firing of numerous input synapses is correlated.Some of these developments may even begin before birth.There is a critical period in early childhood during which such wiring changes may be readily achieved, but some changes in mapping can occur later in life. Some idioms characterize the response properties of neurons (such as the response of many neurons in V1 to orientation), and they are useful.A common term is "feature detector", which really captures the fact that some neurons are sensitive to orientation, others to disparity or wavelength, etc.But it has two disadvantages.First, it implies that the neuron only responds to the "feature" preceding its name. (Some might argue that it's the only neuron responding to that feature, but that's far from the truth.) This ignores the fact that the neuron might also respond to other features (often related ones).For example, an orientation-sensitive cell with an endpoint inhibitory response responds well to short lines (appropriately oriented in place); but due to the substructure of the receptive field, it will also respond to much longer lines partially inside its receptive field The curvature of the straight line is sensitive. A second misconception about feature detectors is that they imply that neurons are used by the brain to generate awareness of that particular feature.That's not necessarily true, for example, a neuron that responds differently to different wavelengths isn't necessarily a central part of the system that makes you see colors.It may belong to another system that merely directs the brain's attention to color differences without producing awareness of that color. On the other hand, features encoded by feature detectors are seldom grouped into elegant types as engineers have designed them to be.This is rarely mentioned these days.For example, one would consider a "simple" type of orientation-selecting cell to have its excitatory and inhibitory regions set up in two ways, one symmetrical along the long axis of the receptive field and the other antisymmetric. ① These types do exist, along with many other related but confusing forms of setting.As we will see in Chapter 13, one can expect this outcome to be precisely the evolution of neural networks using intrinsic learning algorithms, not strictly set in advance by the designer. In order to understand the role a neuron plays in the operation of the brain, we need at least to know its receptive field and where its output projects, i.e. all neurons that have synaptic contact with its axon.Terry Sejnowsh of the Salk Institute calls this "projective field", as opposed to the term "receptive field", when discussing "meaning" (of neurons in the brain) The projected field may have played an important role.If a neuron's axon is severed, its activity won't mean much to the brain. Cortical area V2 (visual area 2) is also large.It also has contralateral visual field mapping like V1, the local scale (called the "magnification factor") of the mapping from the macula to the peripheral V1 varies, if that seems unusual for this, then carefully examine the diagram 45 It can be seen that the mapping of the V2 area is even more peculiar, the mapping is basically divided into two parts, corresponding roughly to the upper and lower parts of the contralateral half of the visual field. ① Likewise, the area dedicated to the portion near the macula is larger than the peripheral portion of the visual field. Overall, the general characteristics of neural insensitivity in V2 are roughly the same as those in V1, such as orientation, movement, parallax, and color, but there are also differences.Nearly all neurons in area V2 receive binocular input.Their receptive fields are often larger than those of neurons in V1 and respond in a more refined way.For example, some neurons respond to certain subjective contours.Although some neurons were also found to fire to the subjective contours of line segment endpoints (Fig. 15) in V1 area, neurons sensitive to other types (such as straight line continuous type, see Fig. 2) did appear only in V2 area, while in V1 area was not found.More than one philosopher was surprised to learn that such neurons respond to subjective contours, but we were not surprised.When we clearly see some visual feature (rather than just infer it), there are areas of our brain where neurons fire to them.This may be a good general rule.If so, it would be a very important rule. Cortical V2 area is also block.Using enzymes that can reveal spots in the V1 region, fairly rough streaks can be seen, running roughly perpendicular to the V1/V2 border.The general visual characteristics to which each type of stripe is sensitive are different.It appears that there are several different streams of information passing through the V2 zone.One handles mostly color information, another handles mostly parallax, and so on.Scientists are intrigued by all these details because they are closely related to the precise way in which various neurons in different subregions are classified and how they allow us to see objects.It is important to us that the behavior of neurons be divided into partially separated categories even within a single region, although the degree of clarity of this separation is debated. So far I've only talked about V1 neurons with projections to V2. Are there neurons in the V2 area that project back ① to the V1 area?The answer is that there are almost as many neurons in V2 with backward projections as there are neurons in V1 with forward projections, but with one important difference.The forward projection is mostly concentrated in the 4th layer of the V2 area, and the feedback to the V1 area completely avoids the 4th layer. It was previously thought that there were only three areas of visual cortex, areas 17, 18 and 19.I have described two of these regions in detail, Region V1 (equivalent to Region 17) and Region V2 (part of the earlier defined Region 18).In addition, how many regions are there?Remarkably, at least twenty different visual areas have now been identified, and seven additional areas are partially related to vision.This fact in itself clearly demonstrates the complexity of visual processing.Because the neurons in each area have different sets of inputs and outputs, they behave very differently.Figure 47 is a model of the macaque flattened cortex constructed by David Van Essen (now at the University of Washington in Seattle).Because the cortex bends and folds, the illustration is necessarily distorted. ① To reduce distortion, selective incisions were made on the cortical sheet to obtain a nearly isolated V1 region, which is inserted on the left side of the figure.Compare this figure with Fig. 48, where the landmarks showing cortical folds are omitted, and a number of cortical areas are drawn in corresponding positions, visual areas and those with partial vision are shaded.In macaques, they take up slightly more than half of the total cortex, (remember that monkeys are very visual animals.) This graph is far from conclusive.For example, the upper right area of ​​46 can still be subdivided.Many regions have odd names, but they are often abbreviations of their full names, such as MT for middle temporal, VIP for ventral intra pari-etal, and so on.Some other areas have numerical designations (omitted here), they are generally defined by Brodman, and some of them have been subdivided (such as 7a and 7b). I will briefly describe two of these areas: MT and V4, so I will not describe everything that is known about all visual areas.This is especially due to the fact that knowledge of many visual areas is still rather poor.The MT area of ​​the cortex is relatively small and is sometimes called the V5 area.It has a fairly good correspondence between visual field half-fields and retinal areas, but the receptive fields of its neurons are generally larger than those of V1 or V2 areas. The neurons in the MT area are particularly sensitive to the movement of the stimulus (including the direction of the movement), and each neuron fires to the stimulus within a certain speed range.Some emit best for high-speed movements, others correspond to low-speed movements. It was not initially thought that the responses of these neurons often depend on the relative motion of the object and the background.John of Caltech.John Allman realizes this.Because, unlike many neuroscientists, he was so interested in monkeys and their wild way of life that he still keeps them at home.He has traveled abroad several times to study monkeys in their natural habitat.He thus had first-hand knowledge of the typical visual environment of monkeys.He attempted to reproduce this environment in a greatly simplified form in the laboratory.He and his colleagues used as stimuli a rod of random dots on a TV screen, and typically a neuron might respond well to a rod of blobs moving up (or down) perpendicular to its length in its receptive field. .However, he found that if the background of blobs was also moving in the same direction, the neuron's firing dropped.If the background was moving in the opposite direction, then the neuron's firing to the moving stick would increase.In this way, neurons mainly detect the relative motion between local features and similar features in the adjacent background.This is the simplest form of the aforementioned non-classical receptive field.Although things are not always so clear-cut,1 it appears that such ensembles of neurons can learn to respond not only to one feature of an object, but also to certain features of the object's environment. Certain neurons in the MT respond to more complex patterns of movement.Their behavior is related to the so-called pinhole problem. Consider Figure 49. Imagine a small circular hole on a screen through which a featureless straight line is observed, which is part of a very long straight line. Most of the long line is hidden by the screen, and if the line moves in any direction, all you can see through the hole is a small line moving perpendicular to its length.This is explained in more detail in the notes to Figure 49. Such is the behavior of neurons in area V1 that are sensitive to the direction of motion.All it can feel is the motion component perpendicular to the line, not the real motion of the entire object.However, some neurons in the MT area do respond to actual movement, especially if the signal consists of a collection of several line segments.Experiments have shown that the neurons in the MT area can be simply divided into two types, one can solve the small hole problem, and the other cannot, just like the neurons in our area.If that's the case, that's great.The truth is much more complicated.Neurons exhibit a wide range of behaviors across the spectrum between these two classes.Nonetheless, this gives an example of how the responses of neurons at higher layers of the visual system can become more refined. If the input is misinterpreted, the brain interprets it incorrectly.A familiar example is the illusion created by the barbershop's rotating column sign—the column actually rotates around its long axis, but the stripes appear to travel upwards in the direction of the column.The actual movement direction of any point on the border of the red and white stripes is perpendicular to the length direction of the column.But the brain sees the stripes moving in the direction of the pillars.Figure 50 explains this phenomenon. Neurons in the MT area of ​​the cortex are barely sensitive to color.Some of them, however, react to the movement of borders formed by only color differences of the same intensity.This is in stark contrast to neurons in the V4 area of ​​the cortex. Neurons in area V4 respond complexly to wavelengths but are barely sensitive to motion. ② Their receptive fields are usually large, but in some cases neurons can respond to small objects with appropriate visual characteristics at any position in the receptive field.This map has a complex correspondence of retinal regions, but not as simple as the V1 area. Many color responses are "double antagonistic responses" that color vision theory leads us to expect.What's more, neurophysiologist Semir Zeki of University College London has shown that their behavior has a Rand effect (see Chapter 4).Their responses depend not only on the wavelengths of light in the center and periphery of the receptive field, but are also strongly influenced by the wavelengths of light from adjacent surfaces.Roughly speaking, they didn't just respond to wavelengths; they responded to perceived color, with one neuron in macaque V4 responding to a red patch in a pattern of differently colored rectangles.And Zeki himself thought it was red.Even if there is interference of the wavelength of the illuminating light, the actual wavelength of the light reaching the retina from the color patch is very different, and the neuron can still respond.This is clearly another example of the environment influencing the behavior of neurons.It is important for psychologists to realize that responses to the environment are to some extent exclusively expressed by individual neurons; they should take this into account in their theoretical models. Figure 48 gives a schematic diagram of the currently known visual areas, but does not involve the connection between them.In general, the main flow of information begins in the V1 area of ​​the cortex on the left and flows to those areas on the far right near the junction of the front of the brain with the non-visual areas of the cortex.These projections are usually roughly represented by a rough map, which means that axon terminals that are close to each other in the receptive area generally come from neurons that are not too far apart in the sending area.This also occurs in regions that do not have a retinal counterpart, such as regions higher up in the hierarchy. Van Essen and colleagues tried to adopt an idea first proposed by the neuroanatomists Kathleen Rockland and Deepak Pandya to arrange all visual areas into a rough hierarchy.Rockland and Pandya specifically pointed out that if the projection from area A to area B is concentrated in layer 4, then the feedback from B to A generally avoids layer 4 and usually has a strong connection with layer 1.We've seen this happen with connections between V1 and V2.This view can be represented quite simply, as shown in Figure 51.The projection from the eye to the brain (mainly concentrated on layer 4) is called "forward projection", and the opposite direction is called "reverse projection". Does this rule about layer 4 connections always hold?The truth is more complicated.It has been shown, however, that, using the convention of Figure 51, it is possible to represent most of the known connections in a single hierarchical graph.One of the latest forms is shown in Figure 52. (Don't forget that each line in the diagram represents a large number of axons in both directions.) You don't have to be intimidated by the intricate details of this connection diagram, just note that it captures the complexity of visual processing (if you can't see other things).Very few people think that their brains are structured in this way. There are some exceptions to the agreement on Layer 4 conventions that are worth noting.For example there are many interconnections between cortical areas at the same level.Simple layer 4 rules don't cover them.Thus more elaborate rules are used in constructing the graph.It is not yet clear whether the real layouts are merely quasi-hierarchical, or whether exceptions to these more complex rules are mainly due to experimental error, but there is no doubt that regions can be roughly arranged in an approximate hierarchy.Does the exception, if any, have special significance?Only further work can answer this question.Area.All connections are bidirectional, and this rule is almost always true, but there are exceptions. ①By the way, Figure 52 does not intend to show connection strength (for example, how many axons each straight line represents), mainly because there is too little information in this regard.Some lines in Figure 52 represent millions of axons, others may have as few as a hundred thousand, or even fewer. Are adjacent regions in the cortex always connected to each other?Usually this is the case, but there are a few exceptions. Rankings are also supported by evidence from different sources.It is a general law of neuron activity in different regions. As we go up the level, its behavior roughly follows two laws: the size of the receptive field increases continuously, so the receptive field in the highest layer usually covers the entire half of the visual field, It even partially or fully includes the other half of the visual field (this is mainly achieved through the corpus callosum connection).In addition, the features that elicit neuronal responses become increasingly complex. Some neurons in the V2 area responded to certain subjective contours, while some neurons in the MT area responded to slightly simpler motion patterns (we have seen that they are able to resolve or partially resolve the pinhole problem). The neurons in the MST area respond to the movement in the entire field of vision, some of which correspond to the object being gradually approaching and becoming larger, and some correspond to the object moving backward, and the neurons in the V4 area respond to color perception, not only Just the wavelength of light. In the higher cortex, we found neurons that responded to the front of the face.It is not sensitive to the position of the face relative to the center of gaze, even when the face is slightly tilted.Such neurons barely respond to random combinations of eyes, noses, mouths, and so on.Other neurons were most sensitive to the side of the face. On the other hand, neurons in area 7a were primarily sensitive to the position of an object relative to the head or body, and were less concerned with what the object was.The latter is the main task of the inferior temporal lobe (the area with IT in the middle of those abbreviations, such as CITd), which has been mentioned in the description of face recognition.There are almost certainly many more complex reactions to be found. From this, it can be seen that in general each region receives several inputs from lower layer regions. (These low-level regions extract more complex features than the fairly simple features reflected in V1.) It then operates on combinations of these inputs to produce more complex features and passes them on to higher levels in the hierarchy. level.At the same time, information flows up the hierarchy in several interacting streams.We have seen examples such as partially separated M and P signals from the retina, the three streams of information from V1 to V2, and the "what" and "where" at a higher level.It must be emphasized, however, that there is often some exchange of information between these streams. What about the reverse channel?This also urgently requires more detailed research.One can imagine various functions for them.They may help shape the aforementioned non-traditional receptive fields, allowing actions at higher levels to affect lower levels.They may also belong to a higher-level system that signals back to lower-level regions when their operations have succeeded on a somewhat global level that their synapses should be corrected to make it easier in the future. detect this feature.They may also be closely related to mechanisms of attention and visual imagery.They may have a role in synchronizing neural oscillations (see Chapter 17).These are only possibilities, but which of them are facts remains to be further investigated. In addition, the whole system does not look like a fixed reaction device.It's more like governed by many momentary dynamic interactions conducted at fairly high velocities.Finally, let's not forget that everything I've described applies to macaques, not us humans.Of course we have reason to assume that our own visual system is similar to macaques, but this is only an assumption.As far as all of our present knowledge is concerned, the difference may be not only in the details, but also in its complexity. If there's something secret to the neocortex, it's its ability to evolve new levels at the processing scale, especially at those higher levels.These extra layers of processing may be the characteristic that distinguishes humans or higher animals from lower animals like hedgehogs.I suspect that the neocortex uses some special learning algorithm so that although each cortical area is involved in a complex processing hierarchy, each of them can extract new types from experience.This ability may distinguish the cerebral cortex from other forms of neural structures, such as the cerebellum and striatum, which do not have this complex form of hierarchy. These ideas are speculative, but one thing is fairly clear: Although there are many different visual areas, each of which analyzes visual input in different and complex ways, it has so far been impossible to locate a single area whose neural activity corresponds precisely to We see vivid images of the world in front of us. Looking at Figure 52, one might think that all this may happen in some more advanced and complex structures (such as the hippocampus) and related cortical structures (labeled HC and ER).They are at the top of the hierarchy.But as we shall see in Chapter 12, a person can lose all these areas on both sides of the brain and still report that he can see well and behave as if he did.In short, while we know how the brain breaks down visual images, we still don't know how it puts them together, how it constructs a well-organized and detailed visual awareness of all the objects in the field of view and their behavior Woolen cloth? ①The precise pattern of stripes and spots is roughly similar in different monkeys of the same species, but not identical in detail.Even in a monkey, the patterns on one side of the brain were different from the other.It's just as if the fingerprints on your left hand are not exactly the same as your right hand, for the same reason that such details depend somewhat on chance events in development.Once again we are confronted with a form that has a certain order but is markedly disorganized in detail. ①The biggest confusion is whether this kind of cells can complete the Fourier transform of the visual scene.It's literally ridiculous.In any case, they are more suitable for performing Gabor (Cabor) transformation.But whether even this view has practical utility remains to be determined.To be sure, some neurons respond best to fine details (these are often called "spatial frequencies"), while others respond better to intermediate or coarser details. ①如图15所示，它们可能参与形成由直线端点构成的错觉轮廓。 ①前者相应于一个衰减的余弦波，而后者相应于衰减的正弦波。 ①这有助于我们去领会在展平的皮层表面显示凝视中心及视野的水平和垂直子午线的位置的那些标志。 ②主观轮廓，也称作"错觉轮廓"，是我们看到的一些虚假的直线，它们实际上在视野中并不存在（如图2和图15）。 ①我称之为"反向投射"，因为习惯上把从视网膜到侧膝体到V1期后到V2的广泛的信息流认为是"向前的"。人工智能领域的工作者通常用自下而上这个术语来代替"向前的"一词。他们称相反方向的信息流为自上而下的。 ①从数学的角度讲，某些位置的高斯曲率远偏离0。 ①最近，哈佛医学院的理查德·波恩（RichardBorn）和罗杰·图特尔（RogerTootell）显示在果猴MT区有两种类型的神经元，每一种都存在于许多小的柱状簇之中。第一种类型的行为与文中的描述大致相同，第二种类型的神经元，其外周并不抑制反而增强神经元的主要反应。 ①该方向也可能向下，这取决于柱子的旋转方向以及条纹画的方式。 ②V4区很大，事实上，范·埃森把它分成三个子区：V4t、V4d，V74v。 （1）V4向V1的反向投射很强，但从VI到V4的向前投射通常很弱，或者没有。