45 |
|
|
46 |
NO SEMANTIC CORRELATIONS!!!!!!! |
NO SEMANTIC CORRELATIONS!!!!!!! |
47 |
|
|
|
MOTIVATING EXAMPLE NOT YET USER-TESTED |
|
|
|
|
48 |
IMAGE: RENDERING MODES, TEXTURE COORDINATES! |
IMAGE: RENDERING MODES, TEXTURE COORDINATES! |
49 |
|
|
50 |
SPARSE CODING: A TEXTURE CONTAINING BOTH YELLOW TRIANGLE AND |
SPARSE CODING: A TEXTURE CONTAINING BOTH YELLOW TRIANGLE AND |
542 |
|
|
543 |
%XXX: shorten by one half column |
%XXX: shorten by one half column |
544 |
|
|
545 |
In this section, we discuss the methods to generate unique |
In this section, we discuss the principles underlying |
546 |
background textures on an abstract level. |
our algorithm to generate unique |
547 |
|
background textures. |
548 |
To be useful, the unique backgrounds should be easily |
To be useful, the unique backgrounds should be easily |
549 |
distinguishable and recognizable, and should not |
distinguishable and recognizable, and should not |
550 |
significantly impair the reading of black text on top of it. |
significantly impair the reading of black text on top of it. |
558 |
In order to design a distinguishable distribution of textures, |
In order to design a distinguishable distribution of textures, |
559 |
we have to take into account the properties of the human |
we have to take into account the properties of the human |
560 |
visual system. |
visual system. |
|
% The seed for randomly choosing |
|
|
% an easily distinguishable unique background from a |
|
|
% distribution based on |
|
561 |
|
|
562 |
%providing an infinite source of unique backgrounds. |
The simple model of texture perception we use assumes |
|
%generating textures based on seed numbers [identity] |
|
|
|
|
|
The simple model we use here assumes |
|
563 |
that at some point, |
that at some point, |
564 |
the results from the different pre-attentive feature detectors, |
the results from the different pre-attentive feature detectors, |
565 |
such as different shapes and colors, |
such as different shapes and colors, |
571 |
in a simple perceptron-like |
in a simple perceptron-like |
572 |
fashion\cite{rosenblatt62neurodynamics,widrow60adaptive}. |
fashion\cite{rosenblatt62neurodynamics,widrow60adaptive}. |
573 |
This configuration is commonly used in neural computation. |
This configuration is commonly used in neural computation. |
574 |
|
As seen for example in \cite{olson02vstm}, only a limited |
575 |
|
number of different features detected can be grouped into |
576 |
|
objects, ... XXX SO!!! JVK |
577 |
|
|
578 |
\begin{figure} |
\begin{figure} |
579 |
%\fbox{\vbox{\vskip 3in}} |
%\fbox{\vbox{\vskip 3in}} |
660 |
or if there were no curved lines, we would be wasting |
or if there were no curved lines, we would be wasting |
661 |
recognition potential by leaving some elements |
recognition potential by leaving some elements |
662 |
of the feature vector always zero. |
of the feature vector always zero. |
663 |
|
However, the results cited above\cite{olson02vstm} also |
664 |
|
indicate that in {\em one} texture, only a limited range |
665 |
|
of features should be used. |
666 |
|
|
667 |
|
|
668 |
Indeed, the entropy is maximized when the features are distributed |
Generally, the entropy is maximized when the features are distributed |
669 |
independently from each other: |
independently from each other: |
670 |
features orthogonal to human perception |
features orthogonal to human perception |
671 |
(e.g.,~color, direction of fastest luminance change) |
(e.g.,~color, direction of fastest luminance change) |