353 |
the algorithm. |
the algorithm. |
354 |
The visual input is transformed into a feature vector, |
The visual input is transformed into a feature vector, |
355 |
which contains numbers (activation levels) corresponding to |
which contains numbers (activation levels) corresponding to |
356 |
e.g. colors, edges, curves and small patterns. |
e.g.~colors, edges, curves and small patterns. |
357 |
The feature vector is matched against the memorized textures |
The feature vector is matched against the memorized textures |
358 |
with a simple model. The simplicity of the feature matcher |
with a simple model. The simplicity of the feature matcher |
359 |
is the key feature of the model and is used in the text |
is the key feature of the model and is used in the text |
382 |
we can formulate the following points: |
we can formulate the following points: |
383 |
To be distinguishable, \label{secfeaturevec} |
To be distinguishable, \label{secfeaturevec} |
384 |
\begin{itemize} |
\begin{itemize} |
385 |
\item A feature vector for a given texture should be always the |
\item A feature vector for a given texture should always be the |
386 |
same, and not a smudged distribution; |
same. |
387 |
therefore, a repeating texture should be easier to recognize |
A non-repeating texture will always have slightly different |
388 |
than one that does not repeat, even if local features |
regions which result in a smudged distribution; |
389 |
are similar. Our anecdotal observations confirm this. |
a repeating texture should be easier to recognize. |
390 |
|
Our anecdotal observations confirm this. |
391 |
% XXX: Why wouldn't it always be the same? |
% XXX: Why wouldn't it always be the same? |
392 |
% - seeing different parts of the texture? |
% - seeing different parts of the texture? |
393 |
% - ambiguous perception? |
% - ambiguous perception? |