256 |
$N$th order statistics of pixels and elaborate connectivity |
$N$th order statistics of pixels and elaborate connectivity |
257 |
structures of certain micropatterns.XXX |
structures of certain micropatterns.XXX |
258 |
|
|
259 |
Textons \cite{julesz81textons}: elongated blobs, line terminators, |
Statistical modeling of textures as samples from a probability |
260 |
line crossings, etc. |
distribution on a random field as already seen in \cite{julesz62visualpattern} |
261 |
|
in a very simple form. |
262 |
|
The most popualar computational approach is Markov random fields |
263 |
|
\cite{cross83markov, geman84stochastic}, where a texture |
264 |
|
is characterized by its local statistics. |
265 |
|
However, these approaches are often not feasible on large |
266 |
|
large neighborhoods, but essentially work on pixel scale. |
267 |
|
|
268 |
|
Attempt to explain texture perception by the densities of textons |
269 |
|
\cite{julesz81textons}, fundamental texture elements, such as |
270 |
|
elongated blobs, line terminators, line crossings, etc. |
271 |
|
However, the textons are hard to define formally. |
272 |
|
|
273 |
|
Textures are continuous |
274 |
|
|
275 |
Of course, such models are not directly applicable on high-resolution |
Of course, such models are not directly applicable on high-resolution |
276 |
textures; some kind of filtering would be required to obtain the input. |
textures; some kind of filtering would be required to obtain the input. |
278 |
\cite{bergen88earlyvision}.XXX |
\cite{bergen88earlyvision}.XXX |
279 |
|
|
280 |
Filtering based approach, e.g., \cite{heeger95pyramid}. |
Filtering based approach, e.g., \cite{heeger95pyramid}. |
281 |
|
Essentially a bank of linear filters is applied to the texture followed |
282 |
|
by a nonlineary and then another set of filters. |
283 |
|
|
284 |
|
XXX: reviews |
|
Statistical modeling of textures as samples from a probability |
|
|
distribution on a random field as already seen in \cite{julesz62visualpattern} |
|
|
in a very simple form. |
|
|
The most popualar approach is Markov random fields |
|
|
\cite{cross83markov, geman84stochastic}, where a texture |
|
|
is characterized by its local statistics. |
|
285 |
|
|
286 |
\subsection{Focus+Context views} |
\subsection{Focus+Context views} |
287 |
|
|