Digital Signal Processing Reference
In-Depth Information
Fig. 3.9
There will be some
problems (see text) if the
whole image is treated as one
document when using LDA to
discover classes of objects
Fig. 3.10
(
a
) Graphical model of SLDA. (
b
) Add spatial information when designing documents.
Each document is associated with a point (marked in magenta color). These points are densely
placed over the image. If an image patch is close to a document, it has a high probability to be
assigned to that document
With the assumption that if two types of image patches are from the same ob-
ject class, they are not only often in the same images but also close in space, a
Spatial Latent Dirichlet Allocation (SLDA) was proposed in [
11
]. Under SLDA,
the word-document assignment becomes a hidden random variable. There is a gen-
erative procedure to assign words to documents. When visual words are close in
space or time, they have a high probability to be grouped into the same document.
The graphical model SLDA is shown in Fig.
3.10
.The
N
visual words in an image
set are assigned to
M
documents.
d
j
is a hidden variable indicating the document
assignment of visual word
i
. Each document
j
is associated with a hyperparam-
eter
c
j
=(
g
j
,
x
j
,
y
j
)
,where
g
j
is the index of the image where the document is