{"id":4701,"date":"2010-07-07T14:55:15","date_gmt":"2010-07-07T14:55:15","guid":{"rendered":"http:\/\/www.labri.fr\/perso\/barla\/blog\/?p=4701"},"modified":"2012-07-09T16:04:47","modified_gmt":"2012-07-09T16:04:47","slug":"computational-vision","status":"publish","type":"post","link":"https:\/\/www.labri.fr\/perso\/barla\/blog\/?p=4701","title":{"rendered":"Computational Vision"},"content":{"rendered":"<p id=\"top\" \/><em>Mallot<\/em><\/p>\n<h2>Edge detection<\/h2>\n<ul>\n<li>If scale-space is used in a coarse-to-\ufb01ne strategy, the results are very sensitive to noise &#8211; p.86<\/li>\n<li>Lines are, in a certain sense, the derivatives of steps, and for this reason, the two types of edges appear in derivatives of different order &#8211; p.86<\/li>\n<li>If the output of even and odd \ufb01lters is squared and these results added, the result in both cases is unambiguous with the maximum at the position of the step or line edge &#8211; p.88<\/li>\n<li>In the two-dimensional case, the use of the Laplacian operator to de\ufb01ne the locations where steepness is greatest is only an approximation, and it leads to systematic localization errors when edges or lines are curved &#8211; p.91<\/li>\n<\/ul>\n<p style=\"text-align: center;\"><a href=\"https:\/\/www.labri.fr\/perso\/barla\/blog\/wp-content\/uploads\/2010\/07\/p78.png\"><img loading=\"lazy\" decoding=\"async\" class=\"size-medium wp-image-4711 aligncenter\" title=\"p78\" src=\"https:\/\/www.labri.fr\/perso\/barla\/blog\/wp-content\/uploads\/2010\/07\/p78-299x300.png\" alt=\"\" width=\"299\" height=\"300\" \/><\/a><\/p>\n<h2>Stereoscopic Vision<\/h2>\n<ul>\n<li>The angle at which the optical axis of the cameras intersect is called the vergence angle &#8211; p.126<\/li>\n<li>The average of the two directions of view is called the version angle &#8211; p.126<\/li>\n<li>Version and vergence are known as Hering\u2019s coordinates of the plane which passes through the nodal point and the \ufb01xation point &#8211; p.127<\/li>\n<li>The kinematics of the eye movements (slow vergence, fast version) and the unnecessary movements of the left eye suggest strongly that physiological control of vergence and of version is separate &#8211; p.128<\/li>\n<li>By means of appropriate neural network models, it has been demonstrated that the calculation of stereoscopic disparities can be achieved through cooperation and inhibition among disparity-selective neurons &#8211; p.141<\/li>\n<\/ul>\n<h2>Color and color constancy<\/h2>\n<ul>\n<li>Codes with overapping sensitivities are called population codes, since each signal (for color, each wavelength) is coded in a distributed excitation pattern which involves all channels. In contrast to this is interval coding, in which the sensitivities of the channels do not overlap &#8211; p.109<\/li>\n<li>For overlapping sensitivities, a difference in excitation of two channels can encode very small gradations in the overlap region. In interval coding, on the other hand, differences in channel activity contain very little information &#8211; p.110<\/li>\n<li>A \ufb01rst order model of color constancy is that a tristimulus vector which corresponds to the averages stimulus color de\ufb01ned as white, and all other vectors are references to it by adaptation of the sensitivity of the receptor. This description is based on the simplifying assumption that the average color of a scene is gray or white, the &#8220;gray-world assumption&#8221; &#8211; p.111<\/li>\n<li>If the color composition of the scene changes suddenly, an after effect called successive color contrast occurs : color constancy is bound up with an adaptation to the predominant color of a scene &#8211; p.111<\/li>\n<li>The retinex theory is based on the importance of edges for color constancy &#8211; p.115<\/li>\n<li>The retinex theory performs satisfactorily if the goal is to eliminate minor changes in illumination from the image. &#8211; p.117<\/li>\n<\/ul>\n<h2>Shape from shading<\/h2>\n<ul>\n<li>Information about light source position of the light source appears to play a minor role. Correlation between the accuracy of depth estimates and estimates of light source position is very poor &#8211; p.148<\/li>\n<li>Shape constancy for different directions of illumination is also poor, i.e. the same surface appears to have different shapes when illuminated from different directions. &#8211; p.148<\/li>\n<\/ul>\n<h2>Motion detection<\/h2>\n<ul>\n<li>The approaches to motion calculation can be distinguished by the type of sampling which takes place in a spatio-temporal cube of image intensities (time discrete or continuous change in the gray value at \ufb01xed locations) &#8211; p.181<\/li>\n<li>In the case of the discrete-time approach, image features must be followed from one image to the next, which is dif\ufb01cult if the time interval is large. Consequently, a correspondence problem arises &#8211; p.182<\/li>\n<li>If the detector is strictly localized in space (and continuous in time), then the problem arises that the brightness variations at one point can be due to entirely different motions. In this case, the image is being observed, so to speak, through a small opening mask (aperture), and so this effect is called the aperture problem &#8211; p.182<\/li>\n<li>In human motion perception, evidence for tuned, bilocal motion detectors based on correlation has been presented by van Doorn and Koenderink &#8211; p.186<\/li>\n<li>Expected similarities between neighboring displacement vectors may be appied in order better to estimate the displacement \ufb01eld. &#8211; p.191<\/li>\n<li>It can be concluded that the human visual system favors the assumption of smoothness over that of rigidity of motion, even if a rigid interpretation exists &#8211; p.194<\/li>\n<li>The spatio-temporal energy approach has become very important in connection with the neurobiology of, at least, vertebrates, since the motion energy with suitable parameters for the \ufb01lter functions, accurately describes the receptive \ufb01elds of motion-selective neurons in the visual cortex &#8211; p.197<\/li>\n<\/ul>\n<p><a href=\"https:\/\/www.labri.fr\/perso\/barla\/blog\/wp-content\/uploads\/2010\/07\/p186.png\"><img loading=\"lazy\" decoding=\"async\" class=\"aligncenter size-medium wp-image-4721\" title=\"p186\" src=\"https:\/\/www.labri.fr\/perso\/barla\/blog\/wp-content\/uploads\/2010\/07\/p186-300x283.png\" alt=\"\" width=\"300\" height=\"283\" srcset=\"https:\/\/www.labri.fr\/perso\/barla\/blog\/wp-content\/uploads\/2010\/07\/p186-300x283.png 300w, https:\/\/www.labri.fr\/perso\/barla\/blog\/wp-content\/uploads\/2010\/07\/p186-768x725.png 768w, https:\/\/www.labri.fr\/perso\/barla\/blog\/wp-content\/uploads\/2010\/07\/p186-1024x967.png 1024w, https:\/\/www.labri.fr\/perso\/barla\/blog\/wp-content\/uploads\/2010\/07\/p186.png 1440w\" sizes=\"auto, (max-width: 300px) 100vw, 300px\" \/><\/a><\/p>\n<h2>Optical \ufb02ow<\/h2>\n<ul>\n<li>It is possible to determine egomotion by calculating a number of \ufb02ow \ufb01elds expected for different types of motion, and then testing which of these corresponds best to the \ufb02ow \ufb01eld actually measured &#8211; p.219<\/li>\n<li>Similar receptive \ufb01elds specialized for speci\ufb01c \ufb02ow patterns are also found in the temporal lobes of the cerebral cortex of primates. These \ufb01elds are, however, markedly smaller and appear, rather, to analyze only parts of the \ufb02ow pattern &#8211; p.220<\/li>\n<\/ul>\n","protected":false},"excerpt":{"rendered":"<p>Mallot Edge detection If scale-space is used in a coarse-to-\ufb01ne strategy, the results are very sensitive to noise &#8211; p.86 Lines are, in a certain sense, the derivatives of steps, and for this reason, the two types of edges appear in derivatives of different order &#8211; p.86 If the output of even and odd \ufb01lters &#8230; <a title=\"Computational Vision\" class=\"read-more\" href=\"https:\/\/www.labri.fr\/perso\/barla\/blog\/?p=4701\" aria-label=\"Read more about Computational Vision\">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-4701","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\/4701","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=4701"}],"version-history":[{"count":4,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts\/4701\/revisions"}],"predecessor-version":[{"id":22841,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=\/wp\/v2\/posts\/4701\/revisions\/22841"}],"wp:attachment":[{"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=4701"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=4701"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.labri.fr\/perso\/barla\/blog\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=4701"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}