Unsolved mystery of view

https://www.sciencedirect.com/science/article/pii/S0960982207014194
  • Transfer
The question of vision is a prominent part of neuroscience. Huge volumes of literature and four Nobel prizes are devoted to this issue, but in the current situation it is impossible not to notice that the mammalian viewpoint described in the textbooks does not cope with the task. The purpose of this essay is to show a set of reasons why you should not close your eyes to it. In fact, a portrait of the mystery of vision will be presented, ranging from a variety of small details at the very beginning of the flow of visual information in mammals, the threat of ignoring them, and ending with a bunch of problems in understanding brain processing at the end of the path.

Device vision system


In the eyes of any textbook on vision, we see in three stages. The first stage: the light hits the retina and is transformed into nervous excitation of photoreceptors - sensory neurons of the retina. In addition, the eye normalizes contrast and brightness, focuses the image.The second stage: the process in the retina, when the map of nervous excitations of photoreceptors is transformed into a parallel information flow, each element of which encodes its own aspect of visual information. The ganglion cells, whose axons form the optic nerve, are much smaller in the retina than the photoreceptors, the first compression of the information occurs. One ganglion cell can be responsible for coding a signal from several in the center to hundreds and thousands of photoreceptors at the periphery of the retina. (Figure 1) Ganglion cells are divided into two main types, depending on which route from the optic nerve signal enters the cortex of the cerebral hemispheres. One type, which is predominantly quantitative (80% in humans), gives a high visual acuity and color perception, for them a stable image contrast is important. For historical reasons, they are called PC cells or dwarf, since the signal from them passes through parvocellular (small cell) layers of lateral geniculate bodies. The second type of MC cells, their signal goes through the magnocellular (macrocellular) layers of the lateral cranked bodies. Responsible for the perception of motion, as they have an increased sensitivity to images that flash or move.






Figure 1. Measuring properties of receptive fields. (A) The reaction of a neuron (a sequence of action potentials or “commissures”) is controlled by an extracellular recording by an electrode in an intermediate cranked body (LGN) of an anesthetized animal. The stimuli are fed to the corresponding area of ​​the neuron field of view. (B) Table of conditional responses of different types of neurons. Each column shows the response of one type of neuron. Each row shows the responses to one type of stimulus. For example, a neuron that is selective in the direction of movement (the second column) only responds when the stimulus moves through the receptive field from bottom to left, and up to the right. Neurons that have standard “center / surroundings” receptive fields are tolerant in size and mobility of the stimulus. Non-standard types of neurons demonstrate much more uncompromising demands on the stimulus.

Both types of these cells have a center / surroundings organization: a supposedly universal coding strategy for visual information. (Figure 2) The best stimulus to achieve the most energetic response from ganglion cells is a more or less round and small spot on the retina, and the more intense it is contrasted with the surrounding background (bright or dark spot), the more vigorous is the response of the cells.

The third stage: the signal is processed in the brain. From the generally accepted point of view, the main processes of processing by the brain begin in the primary visual cortex. Another significant transcoding occurs, neurons do not just react to dark and light spots on the retinal surface, but also become selectively sensitive to oriented contrast boundaries, oriented line segments and their ends. There are also more complex configurations of stimuli, when neurons react precisely to the border of contrast regardless of position, so long as it is in the area of ​​responsibility of the cell. This "complex" processing in the visual cortex is used as the second level of abstraction in the perception of a static picture, where the recognition went to a new, more general level of border perception, regardless of their position. Eventually: first, the retina transmits simple signals to the cerebral cortex; secondly, the neurons of the primary visual cortex use these simple signals to detect contrasting borders and lines; and thirdly, these fundamental building blocks are used to define the boundaries of whole objects and create visual perception.

Probably no one imagined that this point of view would turn into a granite of fundamental dogma. Edge detection has become viewed as if it were the only way to see in mammals, as the evolutionary found, the perfect solution to the problem of effective image perception. In this essay, this point of view will be questioned and uncomfortable questions will be asked about the nature of vision. They have already been raised more than once in the history of the study of vision and do not imply simple answers. So now the main task is to concentrate on these issues, bringing together reasons for doubt in the modern understanding of the device of view.

The retina of all mammals sends non-standard signals to the brain.


A careful study of the route of the visual flow from the eyes to the cerebral cortex in primates categorically confirms that, like in other mammalian species, the functional contribution to the primate visual flow is made by many parallel channels [1–4]. (Figure 2) Some of these channels even ignore the primary visual cortex and immediately go to the higher hierarchy departments [5]. Much of this was mentioned in classical studies of the anatomy of cats and monkeys [6–10], but was ignored by the pioneers of neurobiology in the study of the visual cortex. The aggressive use of Occam's razor principle made it possible to give meaning to the properties of cortical neurons, as they build their function of recognizing lines or borders solely on the basis of inputs from ganglion cells with the center / surroundings organization [11]. Relatively low numbers of primate ganglion cells with non-standard organization were discarded. What is their role in visual perception will be described further, and from the point of view of neurobiology practitioners, it is worth noting that with an arbitrary implantation of an electrode into the optic nerve, the probability of encountering an axon of a non-standard ganglion cell is quite small. Since non-standard cells are rare, most experimenters chose the path of least resistance, creating techniques for projecting visual stimuli onto the retina, best suited for studying center / surroundings, which turned non-standard cells into a statistical “outlier” in experimental data. What is their role in visual perception will be described further, and from the point of view of neurobiology practitioners, it is worth noting that with an arbitrary implantation of an electrode into the optic nerve, the probability of encountering an axon of a non-standard ganglion cell is quite small. Since non-standard cells are rare, most experimenters chose the path of least resistance, creating techniques for projecting visual stimuli onto the retina, best suited for studying center / surroundings, which turned non-standard cells into a statistical “outlier” in experimental data. What is their role in visual perception will be described further, and from the point of view of neurobiology practitioners, it is worth noting that with an arbitrary implantation of an electrode into the optic nerve, the probability of encountering an axon of a non-standard ganglion cell is quite small. Since non-standard cells are rare, most experimenters chose the path of least resistance, creating techniques for projecting visual stimuli onto the retina, best suited for studying center / surroundings, which turned non-standard cells into a statistical “outlier” in experimental data.




Figure 2. Views on the visual system.
(A) Viewpoint from textbooks on the structure of the visual system of primates using the example of a macaque. The retina is filled with a large number of standard ganglion cells, the axons of which form the optic nerve. The retina also contains a small number of nonstandard ganglion cells that respond to light in other ways, but traditionally it is assumed that they are projected only into the midbrain autonomic centers, namely, the upper dvolimie (SC). Concentric cells protrude into a relay nucleus in the thalamus, an intermediate cranial body (LGN), which, in turn, transmits signals to the primary visual cortex (V1). A standard cell through an intermediate crank body (LGN) transmit a signal to the primary visual cortex (V1) and then to the higher sections of the visual cortex (V2, V3, MT).

(B) A more realistic look at the primate visual system. Some of the non-standard ganglion cells poison the signal directly into the intermediate cranial body (LGN), and then into the primary visual cortex. In addition, the upper dvuholmie is not a dead end direction, from there the signal goes into the intermediate cranked body (LGN) and beyond. But this is not all, starting from the intermediate crank body (LGN), the visual flow ceases to be unidirectional, and its reverse component is more powerful than the direct one.


The second problem is illustrated in Figure 3. The top graphic of the figure shows the response of the “center / surround” cell. The basic reaction scheme is very familiar to the neuroscientists involved in vision: the standard center / surroundings ganglion cells show the curve of the response line to the spatial frequency - the maximum response occurs when the lattice band exactly coincides in width with the center section of the center / cell receptive field. environment". But now we will consider the answer of one of the non-standard cells in the optic nerve obtained in the same experiment (Figure 3 b). This time, the cell belongs to the so-called coniocellular type of ganglion cells, a consolidated set of non-standard cell types, which is usually generalized, if it is generally referred to as the “blue path”. This cell has an organization of the type "contrast suppression" [6,7, 12,13], in this experiment, its response is almost completely opposite to the answer of the standard cell "center / environment". When taking into account corrections for differences in the visual acuity of monkeys and cats, the ganglion cells of "contrast suppression" behave the same way [12,13].


Figure 3. Comparison of standard and non-standard receptive field. Cells were stimulated by a drift grating with variable spatial frequency (the frequency here is expressed as the total width of two black and white bands in the grid per degree of viewing angle). The standard “center / environment” ganglion cell (A) behaves as expected: when the width of the lattice band is close to the width of the center of the receptive field, the cell gives a strong response. A non-standard cell with “contrast suppression” (B), instead of excitement, is silent, as long as it is able to distinguish between black and white stripes.

We can only assume that these cells recognize a monotonous fill in their area of ​​responsibility, for example, a clear blue sky, since as the width of the white and black bars of the grid decreases below the resolution of the human eye, they merge into solid gray. A true understanding of the coding of visual information for these cells is not represented in the standard vision model.

All non-reluctant see well using non-standard cells


Advocates of the accepted model of vision when considering primates can refer to a relatively extremely small number of non-standard cells, but this does not apply to retinas of rabbits, cats, rats and mice [7,8,14–16]. In addition, in absolute numbers (about 100,000), the number of non-standard cells in the retina of primates is close to the total number of cells in the retina of a rat or cat [17]. Visual acuity in cats, rabbits, rats and mice is lower than in primates, but acuity is not all. The vision of all these creatures allows you to successfully cope with survival in the wild. Anyone who doubts this is invited to try to sneak up on a rabbit in the field or catch a mouse without using a mousetrap (a simple task for an ordinary cat).


Figure 4. Visual perception with low details.
Observers who are familiar with the personas represented in the image recognize them despite their poor detail. When the image is blurred, enough visual information remains for the person. From left to right: Prince Charles, Woody Allen, Bill Clinton, Saddam Hussein, Richard Nixon, Princess Diana.


It is clear, at least for medical reasons, why neuroscientists have concentrated on the primate retina. But such a selective concentration when trying to understand vision leads to the fact that experimenters introduce and transfer the results of self-perception to the interpretation of the results of the research conducted. The dense packing of primate ganglion cells is well suited for detecting edible fruits in a tree, or for reading the latest issue of the New York Times, but there is no particular reason to consider such tasks as summum bonum (Latin: highest good) of view. In the overall evolutionary picture, this extremely high visual acuity can be viewed as a specific niche adaptation for the detection of high-contrast objects at a great distance. Most mammals in the retina do not have fovea (central fossa in the retina), a specialized area with high visual acuity found in primates, but vision without a million tightly packed ganglion cells remains fully functional. The proof is illustrated in Figure 4, people are easily capable of recognizing without the usual visual acuity [18]. In non-primate mammals, the coding of visual stimuli has been particularly well studied in rabbits [19–23]. Rabbits have large eyes, and the optical system is better than many species of primates, also in their retina there are standard ganglion cells "center / surroundings". However, these standard cells make up only a quarter of the entire population of retinal ganglion cells, represented by about twelve functional and anatomical types [14,21]. Clear, that each of these twelve types of cells is sharpened by evolution under its visual stimuli. But the importance of this fact lies in the fact that the channels from the non-standard ganglion cells dominate in the visual flow of the current into the rabbit brain. This information is not new: innovation is evidence that the retina of all mammals transmits a variety of coding visual channels to the brain [4, 13, 24, 25] and that such diversity should be included in any realistic view of the vision device.

Famous examples are cells that selectively respond to the direction of movement of a stimulus. One type of them directly sends a signal to the nuclei of the brain stem associated with eye movement [26]. But not so well known is another type of these cells, the signal from which goes not only to the midbrain, but also through the lateral geniculate bodies to the primary visual cortex [19]. Such cells are not so rare, but references to them in textbooks when describing a vision device are not found. The signal from nonstandard ganglion cells unambiguously reaches the visual cortex, but how does it use it?

The second non-standard cell type (identified in rabbits, cats, and mice) is the so-called local edge detector. Like cells selectively reacting to the direction of movement, they were also once considered rare ganglion cells, but now we know that this was due to an error when using electrodes: they probably make up about 15% of all ganglion cells [23]. The signals from these cells also reach the primary visual cortex, but their responses lie far beyond the standard visual responses [8,22,23]. These cells respond best to small, slowly moving targets, such as a predator or a predatory animal moving at a distance. However, if there are many small objects, like on a textured field, the cells stop responding. In other words, cells react to a small moving object,

The local edge detector seems to be the most numerous type of rabbit retinal ganglion cells, but it was not included in the standard vision model. Can these cells form a motion analysis system at remote distances? How does the visual cortex process signals from these cells? How do other mammalian species perceive the world with standard ganglion cells and local edge detectors — or even cells with a different compression of visual information that has yet to be discovered?

The visual cortex is smarter than textbooks recognize.


Even if we make the assumption that the entire visual stream is encoded only by standard retinal cells, at this point in time it is known that the processing in the primary visual cortex does not occur solely according to the originally proposed method using “simple”, “complex”, “hypercomplex” neurons, but also neurons "with end braking". Discussion of this fact has been given a lot of space in the specialized literature [27–30].
Problems begin with another obvious fact that almost all neurons in the cerebral cortex are non-linear, which casts doubt on the usefulness of hierarchical schemes with linear assumptions in interpreting their function [31,32].

Secondly, the idea of ​​three main types of cells - “simple”, “complex” and “hypersyllabic” - is a rough abstraction of a rich variety of types of cells of the primary visual cortex, identified by anatomical methods [33,34]. Thirdly, the neurons of the visual cortex depend not only on the incoming visual flow, but also on the different contextual signals of their neighbors, so that the reaction to the same stimuli that are part of the pattern of everyday life and artificially created by experimenters may differ [27,35]. The responses of the visual cortex neurons are not even fixed in time: the removal of a small portion of the retina, which leads to the appearance of a blind spot, allows neurons without work to change their area of ​​responsibility and process visual information from working retina closest to the blind spot [36].

From the point of view of anatomy, none of the above is surprising. Only up to 10% of the input of an individual neuron comes from the channels of the visual flow [37], the rest comes from neighboring neurons or from neurons located in distant superior parts of the cerebral cortex, where visual neurons are still poorly understood and the interpretation of their functions is a difficult task.

Thus, it becomes obvious that the standard view of the device of view is stuck in the swamp of dogma. What steps should be taken to get out of it?

Step 1: Revise the Basics


The latest good news lies in the fact that the methods of visualization of retinal cells over the past 5 years have undergone an inconspicuous revolution, so now it is easy to visualize them in large quantities and with an unprecedented quality of resolution in the form of a three-dimensional image. The problem of finding synapses and the relative position of cells is easily resolved, and now they can be classified not only “by eye”, but also using objective classification methods [15,16]. The degree of conformity of morphological typification to the physiological typification of retinal cells shows a direct connection (discussed in [24]). Thus, the structural types of retinal ganglion cells accurately determine the number of functional channels in the visual stream.

In all mammalian visual systems studied to date, anatomical data indicates that the number of functional channels is approximately twelve. In the retina, monkeys and cats the functions of about half of these channels are well studied. In the rabbit, this proportion is about a third; the mouse and rat retinas, despite their clear advantage for genetic studies, remain largely unexplored.
Thus, it may be useful to return, using modern methods, to the unfinished business of the 1970s [38]. The task is to complete the definition of the functions of the ganglion cells, the first stage of compression of visual information, get rid of the blind spots at the very beginning and update their goals in the rest of the visual system.

Step Fix incentives


Why, after more than half a century of research, is the knowledge of the physiology of vision so far behind the knowledge of anatomy, and what can be done to improve the situation? Pushing aside the problem of the selectivity of research with the help of an electrode, the main problem remains the choice of methodology for finding the right stimulus.

It is much more difficult to understand exactly what retinal ganglion cells and neurons in the primary visual cortex — what is their setup for the characteristics of the visible environment — are used in the vision system — much more difficult than the pioneers of neurobiology thought. The classic technique of the study was that experimenters listened to an amplified signal from an electrode randomly immersed in the optic nerve or in the cerebral cortex while manually moving the stimulus projected on the retina in search of the responsibility area of ​​the cell to which the electrode touched. Fast and effective method, but suffering from subjectivity and lack of reproducibility. Simple trellis stimuli and linear system analysis are extremely effective for standard cells [38,39], but, as mentioned earlier, This method is poorly suited to the study of non-standard cells and nonlinear neurons of the primary visual cortex. There are interesting new attempts to identify the right incentive sets [40], but there is no consensus yet.

The alternative is based on the inverse correlation strategy [41–43], where a stimulus randomly chosen by the experimenter is presented many times, and the cell response is used to calculate the inverse averaging. Thus, the experimenter can construct a representation of the averaged stimulus leading to the most active response of the cell. The great advantage of this method is that it does not imply any particular tuning of the cell to the test stimulus. For both theoretical and analytical reasons, the test stimulus is usually chosen as “random”: a flickering chessboard or some other form of visual noise. This strategy is elegant in terms of concept, but is largely limited to cell analysis with standard center / surroundings,

Another alternative is using a more radical approach. The strategy is to search for the inverse correlation of the cell response to the image, captured on video in everyday situations and presented on the monitor screen [27,44]. The idea is that the researcher is waiting for a signal from the cell when an important stimulus appears. The efficiency of such a method has been proven experimentally, but the question remains as to how natural a two-dimensional image should be considered, how this method will take into account the effects of nonlinearity and context, and even more so how to interpret the results.

New general theory of vision?


Reverse mapping of the surrounding world to signals in the brain can be viewed as empiricism, taken to the extreme. But what can theorists from neuroscience offer in opposition to their fellow experimenters? They have the opportunity to successfully go beyond the currently popular style, which from a formal point of view is only a reformulation of experimental results. Despite a quarter of a century of effort, most of the work done was not able to pass the test of time and did not find confirmation in the new experimental facts. Although accurate and compact formulations of experimental results are important, they rarely lead to a synthesis of new knowledge, and the desire for convenience of mathematical evaluation of data may impose invisible restrictions on the experiments being conducted. A specific example was given earlier:

What, then, can theorists help? They should deal with the inconvenient from the computational point of view aspects of the real nervous system. For example, experimental physiologists know too well that sensory systems are linear only when the experimenter forces them to be such [27,38,45]. A broad coverage of the theory of vision is needed to cope with such tasks as understanding the merits of redundant and sparse coding in visual systems, the need to stretch bridges between studies of vision in invertebrates and vertebrates [46,47]. The question is whether the result will be a rethinking of the first stage in the compression of visual information in the retina [46,48–50]. Previously, the diversity of functional channels of the visual flow was emphasized, which, with all its diversity, serves us in the form of such a familiar feeling - vision.

References
1. Casagrande V.A. (1994). A third parallel visual pathway to primate area V1. Trends Neurosci. 17, 305–310.
2. Hendry S.H.C., and Reid R.C. (2000). The koniocellular pathway in primate vision. Ann. Rev. Neurosci. 23, 127–153.
3. White A.J.R., Solomon S.G., and Martin P.R. (2001). Spatial properties of koniocellular cells in the lateral geniculate nucleus of the marmoset Callithrix jacchus. J. Physiol. 533, 519–535.
4. Dacey D.M., Peterson B.B., Robinson F.R., and Gamlin P.D. (2003). Fireworks in the primate retina: in vitro photodynamics reveals diverse LGN-projecting ganglion cell types. Neuron 37, 15–27.
5. Sincich L.C., Park K.F., Wohlgemuth M.J., and Horton J.C. (2004). Bypassing V1: a direct geniculate input to area MT. Nat. Neurosci. 7, 1123–1128.
6. Rodieck R.W. (1967). Receptive fields in the cat retina: a new type. Science 157, 90–92.
7. Stone J. and Hoffmann K.-P. (1972). Very slow-conducting ganglion cells in the cat’s retina: a major, new functional type? Brain Res. 43, 610–616.
8. Cleland B.G., Levick W.R., Morstyn R., and Wagner H.G. (1976). Lateral geniculate relay of slowly-conducting retinal afferents to cat visual cortex. J. Physiol. 255, 299–320.
9. DeMonasterio F.M. (1978). Properties of ganglion cells with atypical receptive-field organization in retina of macaques. J. Neurophysiol. 41, 1435–1449.
10. Yukie M., and Iwai E. (1981). Direct projection from the dorsal lateral geniculate nucleus to the prestriate cortex in macaque monkeys. J. Comp. Neurol. 201, 81–97.
11. Hubel D.H., and Wiesel T.N. (1977). Ferrier lecture. Functional architecture of macaque monkey visual cortex. Proc. R. Soc. Lond. Ser. B Biol. Sci. 198, 1–59.
12. Troy J.B., Einstein G., Schuurmans R.P., Robson J.G., and Enroth-Cugell C.H. (1989). Responses to sinusoidal gratings of two types of very nonlinear retinal ganglion cells of cat. Visual Neurosci. 3, 213–223.
13. Tailby C., Solomon S.G., Dhruv N.T., Majaj N.J., Sokol S.H., and Lennie P. (2007). A new code for contrast in the primate visual pathway. J. Neurosci. 27, 3904–3909.
14. Rockhill R.L., Daly F.J., MacNeil M.A., Brown S.P., and Masland R.H. (2002). The Diversity of ganglion cells in a mammalian retina. J. Neurosci. 22, 3831–3843.
15. Badea T.C., and Nathans J. (2004). Quantitative analysis of neuronal morphologies in the mouse retina visualized by using a genetically directed reporter. J. Comp. Neurol. 480, 331–351.
16. Kong J.H., Fish D.R., Rockhill R.L., and Masland R.H. (2005). Diversity of ganglion cells in the mouse retina: Unsupervised morphological classification and its limits. J. Comp. Neurol. 489, 293–310.
17. Wassle H. (2004). Parallel processing in the mammalian retina. Nat. Rev. Neurosci. 5, 747–757.
18. Sinha P. (2002). Recognizing complex patterns. Nat. Neurosci. 5 Suppl, 1093–1097.
19. Levick W.R., Oyster C.W., and Takahashi E. (1969). Rabbit lateral geniculate nucleus: sharpener of directional information. Science 165, 712–714.
20. DeVries S.H., and Baylor D.A. (1997). Mosaic arrangement of ganglion cell receptive fields in rabbit retina. J. Neurophysiol. 78, 2048–2060.
21. Roska B., and Werblin F. (2001). Vertical interactions across ten parallel, stacked representations in the mammalian retina. Nature 410, 583–587.
22. Zeck G.M., Xiao Q., and Masland R.H. (2005). The spatial filtering properties of local edge detectors and brisk-sustained retinal ganglion cells. Eur. J. Neurosci. 22, 2016–2026.
23. van Wyk M., Taylor W.R., and Vaney D.I. (2006). Local edge detectors: a substrate for fine spatial vision at low temporal frequencies in rabbit retina. J. Neurosci. 26, 13250–13263.
24. Masland R.H. (2001). Neuronal diversity in the retina. Curr. Opin. Neurobiol. 11, 431–436.
25. Dacey D.M., Liao H.W., Peterson B.B., Robinson F.R., Smith V.C., Pokorny J., Yau K.W., and Gamlin P.D. (2005). Melanopsinexpressing ganglion cells in primate retina signal colour and irradiance and project to the LGN. Nature 433, 749–754.
26. Simpson J.I. (1984). The accessory optic system. Ann. Rev. Neurosci. 7, 13–41.
27. Carandini M., Demb J.B., Mante V., Tolhurst D.J., Dan Y., Olshausen B.A., Gallant J.L., and Rust N.C. (2005). Do we know what the early visual system does? J. Neurosci. 25, 10577–10597.
28. Olshausen B.A., and Field D.J. (2004). Sparse coding of sensory inputs. Curr. Opin Neurobiol. 14, 481–487.
29. Simoncelli E.P., and Olshausen B.A. (2001). Natural image statistics and neural representation. Ann. Rev. Neurosci. 24, 1193–1216.
30. Olshausen B.A., and Field D.J. (2005). How close are we to understanding v1? Neural. Comput. 17, 1665–1699.
31. Ferster D., and Miller K.D. (2000). Neural mechanisms of orientation selectivity in the visual cortex. Ann. Rev. Neurosci. 23, 441–471.
32. Anderson J.S., Lampl I., Gillespie D.C., and Ferster D. (2001). Membrane potential and conductance changes underlying length tuning of cells in cat primary visual cortex. J. Neurosci. 21, 2104–2112.
33. Lund J.S., and Yoshioka T. (1991). Local circuit neurons of macaque monkey striate cortex: III. Neurons of laminae 4B, 4A, and 3B. J. Comp. Neurol. 311, 234–258.
34. Cauli B., Audinat E., Lambolez B., Angulo M.C., Ropert N., Tsuzuki K., Hestrin S., and Rossier J. (1997). Molecular and physiological diversity of cortical nonpyramidal cells. J. Neurosci. 17, 3894–3906.
35. Zipser K., Lamme V.A., and Schiller P.H. (1996). Contextual modulation in primary visual cortex. J. Neurosci. 16, 7376-7389.
36. Gilbert C.D., Sigman M., and Crist R.E. (2001). The neural basis of perceptual learning. Neuron 31, 681–697.
37. Douglas R.J., and Martin K.A. (2004). Neuronal circuits of the neocortex. Ann. Rev. Neurosci. 27, 419–451.
38. Troy J.B., and Shou T. (2002). The receptive fields of cat retinal ganglion cells in physiological and pathological states: where we are after half a century of research. Prog. Ret. Eye Res. 21, 263–302.
39. Lennie P., and Movshon J.A. (2005). Coding of color and form in the geniculostriate visual pathway. J. Opt. Soc. Am. 22, 2013–2033.
40. Victor J.D. (2005). Analyzing receptive fields, classification images and functional images: challenges with opportunities for synergy. Nat. Neurosci. 8, 1651–1656.
41. Srinivasan M.V., Jin Z.F., Stange G., and Ibbotson M.R. (1993). ‘Vector white noise’: a technique for mapping the motion receptive fields of direction-selective visual neurons. Biol. Cybern. 68, 199–207.
42. Meister M., and Berry M.J. (1999). The neural code of the retina. Neuron 22, 435–450.
43. Kara P., Reinagel P., and Reid R.C. (2000). Low response variability in simultaneously recorded retinal, thalamic, and cortical neurons. Neuron 27, 635–646.
44. Ringach D.L., Hawken M.J., and Shapley R. (2002). Receptive field structure of neurons in monkey primary visual cortex revealed by stimulation with natural image sequences. J. Vision 2, 12–24.
45. Schwartz O., and Simoncelli E.P. (2001). Natural signal statistics and sensory gain control. Nat. Neurosci. 4, 819–825.
46. Horridge A. (2000). Seven experiments on pattern vision of the honeybee, with a model. Vision Res. 40, 2589-2603.
47. Srinivasan M.V. (2006). Honeybee vision: in good shape for shape recognition. Curr. Biol. 16, R58–60.
48. Marr D. (1982). Vision: A Computational Investigation in the Human Representation and Processing of Visual Information (New York: W. H. Freeman).
49. Barlow H.B. (1997). The knowledge used in vision and where it comes from. Phil.Trans. Roy. Soc. Lond. B 352, 1141–1147.
50. Riesenhuber M., and Poggio T. (2000). Models of object recognition. Nat. Neurosci. 3, 1199–1204.

Also popular now: