/[gzz]/manuscripts/Paper/paper.tex
ViewVC logotype

Diff of /manuscripts/Paper/paper.tex

Parent Directory Parent Directory | Revision Log Revision Log | View Patch Patch

revision 1.37 by jvk, Thu Mar 20 13:34:11 2003 UTC revision 1.38 by tjl, Thu Mar 20 13:35:26 2003 UTC
# Line 11  Line 11 
11    
12  \newif\ifpics  \newif\ifpics
13  \picsfalse  \picsfalse
14  %\picstrue  \picstrue
15    
16  \title{Rendering recognizably unique textures}  \title{Rendering recognizably unique textures}
17  % Representing Identity  % Representing Identity
# Line 49  NO SEMANTIC CORRELATIONS!!!!!!! Line 49  NO SEMANTIC CORRELATIONS!!!!!!!
49    
50  BLEACHING PIC + ZOOM EFFECT ON READABILITY  BLEACHING PIC + ZOOM EFFECT ON READABILITY
51    
52    MOTIVATING EXAMPLE NOT YET USER-TESTED
53    
54    IMAGE OF INVERTING THE VISUAL SYSTEM!!!
55    
56    IMAGE: RENDERING MODES, TEXTURE COORDINATES!
57    
58    SPARSE CODING: A TEXTURE CONTAINING BOTH YELLOW TRIANGLE AND
59    RED SQUARE MIXES WITH ONE CONTAINING RED TRIANGLE AND YELLOW SQUARE
60    
61  TJL  TJL
62    
# Line 414  c)\\ Line 422  c)\\
422  \includegraphics[width=\fw]{buoyoing.14}  \includegraphics[width=\fw]{buoyoing.14}
423  \includegraphics[width=\fkw]{buoyoing.15}  \includegraphics[width=\fkw]{buoyoing.15}
424  \caption{  \caption{
425  An example of the structure used by BuoyOING.  The motivating example for unique backgrounds:
426  a) A small network of documents.  a focus+context interface for browsing bidirectionally hyperlinked documents.
427  b)..e) The animation seen when traversing the link from node F to H.  The interface shows the relevant {\em fragments} of the other ends of the links
428  In b) we are in the node F and see the relevant {\em fragment} of H,  and animates them fluidly to the focus upon traversing the link.,
429  and in the c) and d) the view fluidly animates to the opposite case.  The (trivial) document network shown in a).
430    The overall organization of the small network used as an example is shown in a).
431    In b) and c) the same sequence of user's views to the network is shown, in b)
432    without and in c) with background texture.
433    There are three keyframes where the view stops and two frames of each animation between the keyframes
434    are shown.
435    The unique backgrounds help the user notice that the upper right buoy in the last keyframe
436    is actually a part of the same document (1) which was in the focus in the first keyframe.
437    Our hypothesis is that this will aid user orientation.
438  }  }
439  \end{figure*}  \end{figure*}
440    
441  In the diagram above, the letters and colors helped identify the documents.  In the diagram above, the letters and colors helped identify the documents.
442  Now,  Now,
443    
444  \section{Unique Background Textures}  \section{Generating Unique Background Textures}
445    
446  %XXX: shorten by one half column  %XXX: shorten by one half column
447    
# Line 434  XXX: - visual discrimination experiments Line 450  XXX: - visual discrimination experiments
450  XXX: simple models (filtering) can have good explanatory power  XXX: simple models (filtering) can have good explanatory power
451  on texture discrimination\cite{bergen88earlyvision}.  on texture discrimination\cite{bergen88earlyvision}.
452    
453  TJL  In this section, we discuss the methods to generate unique
454    background textures on an abstract level.
455    
456  We define a unique background texture as an easily  To be useful, the unique backgrounds should be easily
457  distinguishable and recognizable texture  distinguishable and recognizable, and should not
458  that doesn't  significantly impair the reading of black text on top of it.
 significantly impair the reading of black text painted on top.  
 In this section, we discuss  
 procedural generation  
 of such textures  
 from a seed number, e.g.,~the hash code of the identity of the object  
 to be textured.  
459    
460  The ability to distinguish a particular texture from a large set  The ability to distinguish a particular texture from a large set
461  depends on the distribution of textures in the set.  depends on the distribution of textures in the set.
462  It is intuitively clear that textures with independently  For instance,
463    it is intuitively clear that textures with independently
464  random texel values would be a very bad choice: all such  random texel values would be a very bad choice: all such
465  textures would look alike.  textures would look alike, being just noise.
466  In order to design a distinguishable distribution of textures,  In order to design a distinguishable distribution of textures,
467  we have to take into account the properties of the human  we have to take into account the properties of the human
468  visual system.  visual system.
# Line 471  On a higher level, the correlations betw Line 483  On a higher level, the correlations betw
483  by forming contours and possibly  by forming contours and possibly
484  other higher-level constructions.  other higher-level constructions.
485  These higher levels are not yet thoroughly understood;  These higher levels are not yet thoroughly understood;
486  theories of structural object perception  some theories
487  (see, e.g., Biederman\cite{biederman87})  (see, e.g., Biederman\cite{biederman87})
488  assume certain primitive shapes whose  assume certain primitive shapes whose
489  structure facilitates recognition.  structure facilitates recognition.
# Line 500  about recognizability. Line 512  about recognizability.
512  % The basic assumption of the model is that an image  % The basic assumption of the model is that an image
513  % is perceived as a set of features  % is perceived as a set of features
514    
515  We make the assumption  The simple model we use here assumes
516  that at some point,  that at some point,
517  the results from the  different feature detectors,  the results from the  different feature detectors,
518  such as local and global shapes and colors,  such as local and global shapes and colors,
519  are combined to form an abstract \emph{feature vector}  are combined to form an abstract \emph{feature vector}
520  (see Fig.~\ref{fig-perceptual}).  (see Fig.~\ref{fig-perceptual}).
521  The feature vector is used to compute which concept the particular  The feature vector is then used to compute
522  input corresponds to, in a simple perceptron-like  which concept the particular
523    input corresponds to by comparing it to memorized models
524    in a simple perceptron-like
525  fashion\cite{rosenblatt62neurodynamics,widrow60adaptive}.  fashion\cite{rosenblatt62neurodynamics,widrow60adaptive}.
526  This configuration is sometimes used in neural computation.  This configuration is commonly used in neural computation.
527    
528  The  This
529  rough, qualitative  rough, qualitative
530  model explains readily why uniformly random texels  model is able to explain why uniformly random texels
531  would not make easily distinguishable patterns: different instances  do not make easily distinguishable background textures:
532  of noise would all yield almost  after the ``pre-processing'',
533  exactly the same feature vector in the brain.  different instances
534  Noise has no global shape because there is no correlation between  of noise would all yield
535    {\em almost
536    exactly the same feature vector} in the brain.
537    Noise has no global shapes because there is no correlation between
538  the random local features; it is simply perceived as the distribution  the random local features; it is simply perceived as the distribution
539  of the local features, i.e., color and overall frequency  of the local features, i.e., color and overall frequency
540  (the density of texels).  (the density of texels).
# Line 538  To be distinguishable, \label{secfeature Line 555  To be distinguishable, \label{secfeature
555  % XXX: Why wouldn't it always be the same?  % XXX: Why wouldn't it always be the same?
556  % - seeing different parts of the texture?  % - seeing different parts of the texture?
557  % - ambiguous perception?  % - ambiguous perception?
558  \item There should be as many possible features in the distribution  \item The entropy of the feature vectors
     as possible. For example, if there were no yellow textures,  
     or if there were no curved lines, we would be wasting  
     recognition potential by leaving some elements  
     of the feature vector always zero.  
 \item (Most abstractly)  
     The entropy of the feature vectors  
559      over the distribution of textures, should be maximized.      over the distribution of textures, should be maximized.
560  \end{itemize}  \end{itemize}
561    
# Line 567  To be distinguishable, \label{secfeature Line 578  To be distinguishable, \label{secfeature
578  The last part means essentially  The last part means essentially
579  that if all square-like shapes were green, we would again be  that if all square-like shapes were green, we would again be
580  wasting recognitive power.  wasting recognitive power.
581  Indeed, entropy is maximized when the features are distributed  There should also be as many possible features in the distribution
582        as possible. For example, if there were no yellow textures,
583        or if there were no curved lines, we would be wasting
584        recognition potential by leaving some elements
585        of the feature vector always zero.
586    
587    Indeed, the entropy is maximized when the features are distributed
588  independently from each other:  independently from each other:
589  features orthogonal to human perception  features orthogonal to human perception
590  (e.g.,~color, direction of fastest luminance change)  (e.g.,~color, direction of fastest luminance change)
591  should be independently random, and features not orthogonal  should be independently random, and features not orthogonal
592  (e.g. colors of neighbouring pixels)  (e.g. colors of neighbouring pixels)
593  should be correlated so as to maximize the entropy  should be correlated so as to maximize the entropy.
594  (e.g. pixels on a small area should correlate enough to  For example, pixels on a small area should correlate enough to
595  facilitate perception of contours).  facilitate perception of contours.
596    
597  In a sense, the model of perception should be {\em inverted}  In a sense, the model of perception should be {\em inverted}
598  in order to produce a unique background from  in order to produce a unique background from
# Line 613  in the texture ``channel''. Line 630  in the texture ``channel''.
630    
631  \section{Hardware-accelerated implementation}  \section{Hardware-accelerated implementation}
632    
 TJL  
   
633  In this section, we discuss our hardware-accelerated implementation  In this section, we discuss our hardware-accelerated implementation
634  (libpaper)  (libpaper)
635  of unique backgrounds (papers).  of unique backgrounds (papers).
# Line 1275  as opposed to 2 and 2 on NV10, Line 1290  as opposed to 2 and 2 on NV10,
1290  using 2 passes as we currently do is too much;  using 2 passes as we currently do is too much;
1291  it should be possible to obtain interesting textures with just one pass.  it should be possible to obtain interesting textures with just one pass.
1292  We are also working on implementing  We are also working on implementing
1293  these algorithms on ATI's extensions, due to their recent release  these algorithms on OpenGL ARB extensions ..., due to their recent release
1294  of a Linux driver.  of a Linux driver.
1295    
1296  % However, we see the proprietary extensions only  % However, we see the proprietary extensions only
# Line 1345  Marketta Niemelä and Pertti Saariluoma f Line 1360  Marketta Niemelä and Pertti Saariluoma f
1360  Benja Fallenstein,  Benja Fallenstein,
1361  Matti Katila,  Matti Katila,
1362  and Asko Soukka  and Asko Soukka
1363  have been involved in the development of other aspects of the BuoyOING  have contributed to the development of the BuoyOING
1364  interface (not related to the background textures presented here).  interface in aspects not related to the unique background textures.
1365    
1366    
1367  \bibliographystyle{plain}  \bibliographystyle{plain}

Legend:
Removed from v.1.37  
changed lines
  Added in v.1.38

savannah-hackers-public@gnu.org
ViewVC Help
Powered by ViewVC 1.1.26