[Gzz-commits] manuscripts/Paper paper.tex

[Janne V. Kujala]

CVSROOT:        /cvsroot/gzz
Module name:    manuscripts
Changes by:     Janne V. Kujala <address@hidden>        03/03/20 06:52:29

Modified files:
        Paper          : paper.tex 

Log message:
        reorg

CVSWeb URLs:
http://savannah.gnu.org/cgi-bin/viewcvs/gzz/manuscripts/Paper/paper.tex.diff?tr1=1.34&tr2=1.35&r1=text&r2=text

Patches:
Index: manuscripts/Paper/paper.tex
diff -u manuscripts/Paper/paper.tex:1.34 manuscripts/Paper/paper.tex:1.35
--- manuscripts/Paper/paper.tex:1.34    Wed Mar 19 05:25:58 2003
+++ manuscripts/Paper/paper.tex Thu Mar 20 06:52:29 2003
@@ -1045,6 +1045,51 @@
 %  subjective preference for text displays and empirical measures of their
 %  readability."
 
+It can be argued that
+the backgrounds clutter the display
+visually, making the user interface more confusing,
+and reduce text readability.
+Indeed, one of the most difficult aspects of the work was making the random
+color selection produce acceptable results.
+However, by tuning the color selection and the gamma
+correction of the display, we were able to (in our opinion) avoid
+the above problems.
+It is important that the colors chosen are light and that the palettes
+have a relatively small range of colors.
+Also, text readability on the generated backgrounds depends 
+strongly on the text scale; making it easy for the user to zoom
+fluidly in and out helps.
+
+We have not tried to attain infinite zoomability\cite{furnas00infinity}
+with the current implementation,
+but only the more modest goal of zooming within a range 
+that would be reasonable for a single PDF document, 
+i.e., approximately 100-fold
+difference between mininum and maximum zoom. 
+% This is in line with our
+% intended use, since it would be unreasonable 
+% to expect a texture to be recognizable
+% if a sub-pixel area were zoomed to the full screen.
+It could be possible\cite{furnas00infinity} 
+to make the unique background look similar 
+at different scales, but
+this would remove the use of the texture as a cue of scale.
+Our
+nonlinear use of the register combiners
+does have some ill effects when zooming the texture out
+to a very small scale: mipmapping will not give the correct
+average color value.
+It may be possible to alleviate this by modeling the texture mathematically
+and calculating the correct average and placing corrective terms to the
+equations.
+However, in the intended zooming range 
+the current system is quite satisfactory.
+
+% It is important that the background can be zoomed 
+% to different resolutions.
+
+% leads to aliasing: .... modeling textures mathematically , ...
+
 TJL
 
 The most commonly asked question about this work concerns 
@@ -1070,80 +1115,54 @@
 
 JVK
 
-There has been
-lot of texture perception work on texture discrimination.
-However, in our application texture discrimination is not as
-much of an issue as memorizability and recognizability of
-previously seen textures.
-Furthermore, our textures are on a higher level,
-more like complete pictures than the usually studied microstructure.
-
-
-XXX
-
-%\subsection{Texture perception}
-
 Psychophysical studies on texture perception have mostly concentrated
-on \emph{texture discrimination}, the ability of human observers to
-discriminate pairs of textures.  
-The term is often used interchangably with \emph{texture segregation},
-the more specific task of finding the border between differently textured 
-areas (different phases of local characteristics at the
-border can segregate otherwise indiscriminable textures).
-
-First experiments on computer-generated, unnatural textures in the 60s
-\cite{julesz62visualpattern} led to proposals of discrimination models
-based on the $N$th-order statistics of textures 
+on \emph{texture discrimination}\cite{julesz62visualpattern}, 
+the ability of human observers to discriminate pairs of textures.  
+%The term is often used interchangably with \emph{texture segregation},
+%the more specific task of finding the border between differently textured 
+%areas (different phases of local characteristics at the
+%border can segregate otherwise indiscriminable textures).
+%
+%First experiments on computer-generated, unnatural textures in the 60s
+%\cite{julesz62visualpattern} led to proposals of discrimination models
+%based on the $N$th-order statistics of textures 
+%(the joint distributions of the values at the corners of a randomly
+%placed (translated) $N$-gon for all different $N$-gons).
+%%and connectivity structures of certain micropatterns.
+%
+First discrimination models were based
+on the $N$th-order statistics of textures 
 (the joint distributions of the values at the corners of a randomly
 placed (translated) $N$-gon for all different $N$-gons).
-%and connectivity structures of certain micropatterns.
+However, the order of similarity in the statistics did not 
+consistently explain discrimination performance, and certain
+distinctive local features were conjectured.
 
-Attempt to explain texture discrimination by the densities of textons
-\cite{julesz81textons}, fundamental texture elements, such as
+Julesz\cite{julesz81textons} proposed that discrimination could be explained
+by the densities of textons, fundamental texture elements, such as
 elongated blobs, line terminators, line crossings, etc.  
 However, the textons are hard to define formally.
 
 Much simpler filtering-based models can explain texture discrimination
 just as well \cite{bergen88earlyvision}.
-Essentially a bank of linear filters is applied to the texture followed
-by a nonlinearity and then another set of filters.
-In \cite{heeger95pyramid}, new textures with appearance similar
-to a given texture are created by matching certain histograms 
-of filter responses.
+In this approach, a bank of linear filters is applied to the texture followed
+by a nonlinearity and then another set of filters to extract densities
+of features (see, e.g., \cite{heeger95pyramid} for an application).
+%In \cite{heeger95pyramid}, new textures with appearance similar
+%to a given texture are created by matching certain histograms 
+%of filter responses.
 
-Mapping texture appearance to an Euclidian texture space
-(see \cite{gurnsey01texturespace} and the references therein):
-in the reported experiments, three dimensions have been sufficient
-to explain most of the variation in the similarity judgements for
-artificial textures. 
-However, the texture stimuli have been somewhat simple 
-(no color, lack of frequency-band interaction, etc.).
-For some natural texture sets (see, e.g., \cite{rao96texturenaming}), 
-three dimensions have also been
-sufficient, but often semantic connections cause the
-similarity to be context-dependant, making it hard to assess the 
-dimensionality.
-% XXX: this is something we should experiment with our textures
-
-XXX: reviews
-
-XXX: physiological knowledge of visual perception
-(see, e.g.,~Bruce et al\cite{bruce96visualperception});
-
-XXX: in most work, texture is considered as the output of a stochastic
+In our application texture discrimination is not as
+much of an issue as memorizability and recognizability of
+previously seen textures.
+Furthermore, in most texture perception work
+texture is considered as the output of a stochastic
 process that produces certain repeating features. 
-Different samples from the process are considered as the same texture.
 The textures created by our algorithm, although repeating, are more like 
-complete images rather than microstructure. 
+complete images than statistical microstructure. 
 Therefore, higher level processes of vision are also involved
 in the perception and recognition.
 
-theories of structural object perception 
-(see, e.g., Biederman\cite{biederman87})
-
-
-XXX
-
 Experiments on black-and-white %(faces,) 
 ink blots, and snow crystals
 \cite{goldstein71visualrecognition} show that
@@ -1182,7 +1201,30 @@
 
 XXX: refs?
 
-XXX: texture set dimensionality studies?
+Mapping texture appearance to an Euclidian texture space
+(see \cite{gurnsey01texturespace} and the references therein):
+in the reported experiments, three dimensions have been sufficient
+to explain most of the variation in the similarity judgements for
+artificial textures. 
+However, the texture stimuli have been somewhat simple 
+(no color, lack of frequency-band interaction, etc.).
+For some natural texture sets (see, e.g., \cite{rao96texturenaming}), 
+three dimensions have also been
+sufficient, but often semantic connections cause the
+similarity to be context-dependant, making it hard to assess the 
+dimensionality.
+% XXX: this is something we should experiment with our textures
+
+XXX: reviews
+
+XXX: physiological knowledge of visual perception
+(see, e.g.,~Bruce et al\cite{bruce96visualperception});
+
+theories of structural object perception 
+(see, e.g., Biederman\cite{biederman87})
+
+
+XXX
 
 %\section{Software availability}
 
@@ -1228,54 +1270,7 @@
 to give, e.g., all academic articles stored on a user's hard
 drive their own background.
 
-\subsection{Problems}
-
-It can be argued that
-the backgrounds clutter the display
-visually, making the user interface more confusing,
-and reduce text readability.
-Indeed, one of the most difficult aspects of the work was making the random
-color selection produce acceptable results.
-However, by tuning the color selection and the gamma
-correction of the display, we were able to (in our opinion) avoid
-the above problems.
-It is important that the colors chosen are light and that the palettes
-have a relatively small range of colors.
-Also, text readability on the generated backgrounds depends 
-strongly on the text scale; making it easy for the user to zoom
-fluidly in and out helps.
-
-We have not tried to attain infinite zoomability\cite{furnas00infinity}
-with the current implementation,
-but only the more modest goal of zooming within a range 
-that would be reasonable for a single PDF document, 
-i.e., approximately 100-fold
-difference between mininum and maximum zoom. 
-% This is in line with our
-% intended use, since it would be unreasonable 
-% to expect a texture to be recognizable
-% if a sub-pixel area were zoomed to the full screen.
-It could be possible\cite{furnas00infinity} 
-to make the unique background look similar 
-at different scales, but
-this would remove the use of the texture as a cue of scale.
-Our
-nonlinear use of the register combiners
-does have some ill effects when zooming the texture out
-to a very small scale: mipmapping will not give the correct
-average color value.
-It may be possible to alleviate this by modeling the texture mathematically
-and calculating the correct average and placing corrective terms to the
-equations.
-However, in the intended zooming range 
-the current system is quite satisfactory.
-
-% It is important that the background can be zoomed 
-% to different resolutions.
-
-% leads to aliasing: .... modeling textures mathematically , ...
-
-\subsection{Further work}
+%\subsection{Further work}
 
 So far, we have concentrated mostly on low-end hardware, and
 have not even tapped the full potential of the NV25 architecture.