{"id":3881,"date":"2010-07-06T14:49:53","date_gmt":"2010-07-06T14:49:53","guid":{"rendered":"http:\/\/www.labri.fr\/perso\/barla\/blog\/?p=3881"},"modified":"2019-08-17T10:31:15","modified_gmt":"2019-08-17T10:31:15","slug":"vision-science","status":"publish","type":"post","link":"https:\/\/www.labri.fr\/perso\/barla\/blog\/?p=3881","title":{"rendered":"~Vision Science"},"content":{"rendered":"<p id=\"top\" \/><em>Palmer<\/em><\/p>\n<h2>An introduction to Vision Science<\/h2>\n<ul>\n<li>Vision is a heuristic process in which inferences are made about the most likely environmental condition that could have produced a given image &#8211; p.23, col.2<\/li>\n<\/ul>\n<h2>Theoretical approaches to vision<\/h2>\n<ul>\n<li>A theory is an internally consistent set of hypotheses or assumptions from which one can derive explanations of known facts and testable predictions of new facts &#8211; p.46, col.2<\/li>\n<li>Ocklam\u2019s razor principle : the best theory is the most parsimonious one, i.e. the theory that can account for the empirical results with the fewest assumtions &#8211; p.47, col.1<\/li>\n<li>From a computer vision point of view, complex information processing systems can be organized in three levels : the computational, algorithmic and implementational levels &#8211; p.71, col.2<\/li>\n<li>The computational level speci\ufb01es what computation needs to be performed and on what information it should be based, without specifying how it is accomplished &#8211; p.72, col.1<\/li>\n<li>The algorithmic level speci\ufb01es how a computation is executed in terms of information processing operations : it implies decisions upon a representation for the input and output information and the construction of a set of processes that will transform the input representation into the output representation in a well-de\ufb01ned manner &#8211; p.72, col.2<\/li>\n<li>The implementational level speci\ufb01es how an algorithm is embodied as a physical process within a physical system. In particular, the same algorithm might be implemented on brains as well as various different kinds of computers &#8211; p.73, col.1<\/li>\n<li>There are three types of computational information processing systems : decomposable, nearly decomposables and undecomposable sytems. A decomposable system is one in which the interactions among components are negligible in comparison with those within components, while a nearly decomposable system is one in which those interactions are weak but not negligible and an undecomposable is one where interactions between components are as strong as those within components &#8211; p.75-76, col.2-1<\/li>\n<li>Decomposition is stopped at the hardware primitive level (e.g. neurons in the brain) : this is where information is embodied, or implemented p.76, col.2<\/li>\n<li>According to D. Marr, there are four stages of visual perception. They are named for the kind of information they represent explicitly ; they are the image-based, surface-based, object-based and category-based stages of perception. At each level, current processing is in\ufb02uenced by prior higher-level interpretations &#8211; p.85, col.1-2 <em>This a bit idealized.<\/em><\/li>\n<\/ul>\n<h2>Color vision : a microcosm of vision science<\/h2>\n<ul>\n<li>Purples, browns and desaturated colors are produced by a mixture of monochromatic (spectral) colors &#8211; p.101, col.2<\/li>\n<li>Lightness constancy is perceived by the means of ratios of intensity across edges which are extrapolated over the whole region &#8211; p.129, col.1<\/li>\n<li>There are two different types of intensity edges : re\ufb02ectance and illumination edges &#8211; p.130, col.1<\/li>\n<li>For making a discrimination between re\ufb02ectance and illumination edges, several heuristics are available, e.g. : fuzziness (illumination edges tend to be fuzzy, whereas re\ufb02ectance edges tend to be sharp), planarity (if depth information tells that two regions are not coplanar, then edges between those regions tend to be perceived as illumination edges), ratio magnitude (illuminance edges can produce much greater changes in luminance than re\ufb02ectance edges), etc &#8211; p.132, col.1-2<\/li>\n<\/ul>\n<h2>Processing image structure<\/h2>\n<ul>\n<li>Line terminations can produce illusory contours &#8211; p.182, col.2<\/li>\n<li>There seems to be four different pathways in the visual system : color, form, stereo (depth) and motion. But the separation is not complete : this is only a vast simpli\ufb01cation &#8211; p.197, col.1-2<\/li>\n<\/ul>\n<h2>Perceiving surfaces oriented in depth<\/h2>\n<ul>\n<li> According to J. J. Gibson, surface perception can be conceived as determined by its distance and orientation, the latter being de\ufb01ned by the slant and tilt of the surface with respect to the viewer\u2019s line of sight &#8211; p.200-201, col.2-1<\/li>\n<li>Shape-from-X approaches represent either depth or surface orientation, so it seems more appropriate to call them depth-from-X or orientation-from-X ; shape-from-X is a great deal more complicated -p.203, col.1<\/li>\n<li>There are many different sources of depth information that can be classi\ufb01ed in many different ways : ocular\/optical, monocular\/binocular, static\/dynamic, absolute\/relative, and quantitative\/qualitative informations -p.203, col.2<\/li>\n<li>Accomodation is only used to perceive depth at small distances. Likewise, convergence is limited to short distances. It varies along with accomodation &#8211; p.204-205, col.2-1<\/li>\n<li>There are four types of surfacic edges : orientation (surfaces in contact), depth (visibility discontinuity), illumination (differences in amount of light) and re\ufb02ectance edges (material properties) &#8211; p.238, col.2<\/li>\n<li>The integration of information sources can be made in three different ways : the dominance of one particular source, a compromise between sources or the interaction of multiple sources &#8211; p.247, col.2<\/li>\n<li>Binocular disparity information can be overriden by monocular depth information. Relative size, position relative to the horizon, occlusion and motion parallax seem to be conbined in a weak fusion (there are no interactions between different sources) to re\ufb01ne the perception of depth &#8211; p.248, col.2<\/li>\n<li>In the modi\ufb01ed weak fusion theoretical position, interactions between channels can occur : they consist in promotions of one channel thanks to another one &#8211; p.249, col.1-2<\/li>\n<\/ul>\n<h2>Organizing objects and scenes<\/h2>\n<ul>\n<li> Classical principles of grouping : proximity\/similarity of position, similarity of color, size, orientation, common fate\/similarity of velocity, symmetry, parallelism, continuity, closure &#8211; p.257<\/li>\n<li>New principles of grouping : synchrony, common region, element connectedness &#8211; p.259<\/li>\n<li>Past experience : if elements have been previously associated in prior viewings, they will tend to be seen as grouped in present situations &#8211; p.266, col.1<\/li>\n<li>Grouping occurs at each of the four Marr\u2019s stages, each level superseding the ones before p.266, col.2<\/li>\n<li>According to Wertheimer, uniform connectedness is the tendency to perceive connected regions of uniform image properties as the initial units of perceptual organisation. It is not always true (e.g. camou\ufb02ages) &#8211; p.268, col.1<\/li>\n<li>Grouping rules can help reconstruct the different parts of a partly occluded object &#8211; p.271<\/li>\n<li>Grouping and parsing might be applied simultaneously &#8211; p.275, col.2<\/li>\n<li>Texture segregation seems closely related both to classical grouping and to region segmentation &#8211; p275, col.2<\/li>\n<li>The visual system has a strong preference to ascribe the contour of just one of its bordering regions (the \ufb01gure) and to perceive the other side as part of a surface extending behind it (the ground) &#8211; p.281<\/li>\n<li>Figure\/ground discrimination factors : surroundedness, size, orientation, contrast, symmetry, convexity, parallelism &#8211; p.282-283<\/li>\n<li>Figure\/ground organisation must operate initially before either parsing or grouping can sensibly take place &#8211; p.283<\/li>\n<li>In attending to \ufb01gures rather than ground, the perceiver is selectively processing the non-accidental features of the visual \ufb01eld &#8211; p.284, col.1<\/li>\n<li>A hole is ground for purposes of de\ufb01ning depth relations and what is material versus open space ; but is \ufb01gure for purposes of describing shape &#8211; p.287, col.1<\/li>\n<li>Non-accidentalness gives an explanation to many of the phenomena of perceptual organisation &#8211; p.299, col.1<\/li>\n<\/ul>\n<h2>Perceiving object properties and parts<\/h2>\n<ul>\n<li> Perceptual experience can be considered a blend of proximal (retinal image) mode and distal (real world) mode &#8211; p.313-314<\/li>\n<li>Under normal conditions, relationally determined size perception converges with distance-based size perception to produce normal size constancy. In many cases, it includes both occular and optical information &#8211; p.320-321<\/li>\n<li>The shape constancy may be closely related with the identity of the object, helped by continous motion, and better achieved for objects with axes\/points of symmetry &#8211; p.331<\/li>\n<li>Orientation constancy is permitted by the sum of image orientation and head orientation &#8211; p.334<\/li>\n<li>Head orientation is detected not only via the vestibular system, but also via kinesthetic feedback and normal vision &#8211; p.335<\/li>\n<li>Position constancy is allowed by image displacement + eyes motion (head position + eyes gaze), but only for static, simple con\ufb01gurations &#8211; p.338-339<\/li>\n<li>The mecanisms underlying perceptual constency are heuristic ; thus the same mechanisms that lead to veridical perception cause substancial illusions in some ecologically unusual cases &#8211; p.343<\/li>\n<li>Perception (together with use) speci\ufb01es the parts of objects for which language provides names &#8211; p.349, col.1<\/li>\n<li>Deep concavity rule : a surface should be divided at places where its surface is maximally curved inward &#8211; p.353, col.2<\/li>\n<li>It is highly probable that the experience of global objects precedes that of local parts &#8211; p.355, col.1<\/li>\n<\/ul>\n<h2>Representing shape and structure<\/h2>\n<ul>\n<li> Objective shape is de\ufb01ned as the spatial structure of an object that does not changes under similarity transformations : translations, rotations, dilations, re\ufb02ections and theirs combinations &#8211; p.364<\/li>\n<li>The invariant feature hypothesis suggests that shape might be represented by the set of properties that are invariant over the similarity group : number of lines, angles, relative orientations, sizes, closedness, connectedness &#8211; p.366, col.1<\/li>\n<li>Transformational alignment \ufb01nd a similarity transformation that brings one object into exact alignment with the other &#8211; p.361, col.1<\/li>\n<li>The object-centered reference frame hypothesis suggests that the coordinate system used in describing each object is somehow \u201dmade to order\u201d for that particular object &#8211; p.370, col.2<\/li>\n<li>Perceived shape equivalence : different axes of symmetry can be aligned with gravity &#8211; p.371, col.1<\/li>\n<li>Figures with good intrinsic axes are matched in different orientations ; whereas amorphous or ambiguous \ufb01gures are not &#8211; p.374, col.1<\/li>\n<li>The selection of a reference orientation uses different stimulii : gravitational orientation, axes of re\ufb02ectional symmetry, axes of elongation, contour orientation, contextual orientation and motion &#8211; p.375-376<\/li>\n<li>In template representations, shape is speci\ufb01ed by the concatenation of receptor cells on which the image of a particular object would fall &#8211; p.377, col.2<\/li>\n<li>The object relative power spectrum (Fourier) representation encodes relative amplitudes, spatial frequencies and orientations &#8211; p.383<\/li>\n<li>Multidimensional\/featural representations de\ufb01ne an object\u2019s perceived shape by the set of its spatial (binary\/continuous) features &#8211; p.385<\/li>\n<li>Structural descriptions are representations that contain explicitly informations about parts and relations between parts &#8211; p.394, col.1<\/li>\n<li>Theory of \ufb01gural goodness related to information theory : good \ufb01gures can be described in fewer bits than bad \ufb01gures &#8211; p.400, col.1<\/li>\n<li>Theory of \ufb01gural goodness related to symmetry subgroups : good \ufb01gures can be described by the subset of spatial transformations that leaves it invariant (with different weights for different transformations) &#8211; p.401, col.2<\/li>\n<li>Structural information theory derives shape descriptions by generating and simplifying perceptual descriptions called codes that are suf\ufb01cient to generate the \ufb01gure &#8211; p.402, col.2<\/li>\n<\/ul>\n<h2>Perceiving function and category<\/h2>\n<ul>\n<li> Two theories to the perception of function : affordances (direct\/unmediated approach) and categorization (indirect\/mediated approach) &#8211; p.409-410<\/li>\n<li>Four components of categorization : object representations, category representations, comparison processes and decision processes &#8211; p413, col.2<\/li>\n<\/ul>\n<h2>Perceiving motion and events<\/h2>\n<ul>\n<li> The visual system is much more sensitive to the motion of one object relative to another than it is to the same object relative to the observer &#8211; p.469, col.2<\/li>\n<li>Paradoxical motion : when there is a clean perception of motion but no global change in the perceived position of the moving objects &#8211; p.470, col.1<\/li>\n<li>Simultaneous motion contrast is caused by lateral inhibition in the visual \ufb01eld : motion in the outer region causes motion detectors that are sensitive to movement in the same direction in the central region to \ufb01re rapidly &#8211; p.471, col.1<\/li>\n<li>The visual system solves the correspondence problem almost exclusively according to distance criteria, even if orientation, shape and size can matter &#8211; p.476, col.1<\/li>\n<li>Two motion systems : short-range (small displacements\/sizes, detect perceptual units) and long-range (large displacements, after binocular vision, \ufb01gure\/ground organization, shape and color analysis, constancy and depth achieved) &#8211; p478<\/li>\n<li>The unique point heuristic is the tendency to extrapolate the unambiguous motion of unique points to other parts of the same object (e.g. barberpole illusion metaphor) &#8211; p.480, col.1<\/li>\n<li>One simple and elegant theory speci\ufb01es how to combine the constraints provided by all local motion analyses into a best estimate of the global motion (e.g. Adelson and Movshon 1982) &#8211; p.486, col.2<\/li>\n<li>As long as the observer has good information about how far away an object is, he or she perceives approximately its real world velocity, presumably by taking into account the distance to the object &#8211; p.488, col.2<\/li>\n<li>The rigidity heuristic : all else being equal, if there is an interpretation in which rigid motion can be perceived, it will be &#8211; p.489, col.2<\/li>\n<li>Unique points appear to play an important role in the veridical perception of rigid motion &#8211; p.490, col.1<\/li>\n<li>If no unique point is present, the visual system is fooled and tends to perceive rigid motion in 3D ( !) or even non-rigid motion ( ! !) &#8211; p.492<\/li>\n<li>It appears that the preference for rigid motion will be expressed perceptually only if there is suf\ufb01cient time for the analog process underlying it to complete the appropriate transformation internally &#8211; p.494, col.2<\/li>\n<li>Common fate may be generalized to include other types of rigid motion &#8211; p.499, col.1<\/li>\n<li>Induced motion : when a small \u201dstationary\u201d object is perceived as moving when it is surrounded by a larger moving object &#8211; p.501, col.2<\/li>\n<li>If the whole environment moves around us, we usually perceive ourselves as moving, even if we are actually stationary &#8211; p.502, col.1<\/li>\n<li>Optic \ufb02ow is the dominant proximal stimulus for normal everyday vision &#8211; p.504, col.2<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Palmer An introduction to Vision Science Vision is a heuristic process in which inferences are made about the most likely environmental condition that could have produced a given image &#8211; p.23, col.2 Theoretical approaches to vision A theory is an internally consistent set of hypotheses or assumptions from which one can derive explanations of known &#8230; <a title=\"~Vision Science\" class=\"read-more\" href=\"https:\/\/www.labri.fr\/perso\/barla\/blog\/?p=3881\" aria-label=\"Read more about ~Vision Science\">Read more<\/a><\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[621],"tags":[],"class_list":["post-3881","post","type-post","status-publish","format-standard","hentry","category-books"],"_links":{"self":[{"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts\/3881","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=3881"}],"version-history":[{"count":5,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts\/3881\/revisions"}],"predecessor-version":[{"id":40140,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts\/3881\/revisions\/40140"}],"wp:attachment":[{"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=3881"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=3881"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=3881"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}