-
Notifications
You must be signed in to change notification settings - Fork 12
/
readme.txt
49 lines (39 loc) · 2.49 KB
/
readme.txt
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
This project extends the mrmath library with a few common image processing techniques:
rgb to matrix conversion and vice versa
Procrustes analysis
Delauny triangulation
Image warping based on a triangulation
Image warping based on thin plate splines
An Active Appearance Models framework including an annotation application and a model builder application
As a precondition for that project you also need to checkout http://code.google.com/p/mrmath/ and put it to the projects search path!
This project heavily depends on the functionality provided in
the mrmath library. You can download the libarary from
http://www.mrsoft.org/
or if you like a convenient svn download then check out
https://github.com/mikerabat/mrmath
You can add the matrix library to your projects search path for easy
integration.
##############################################################
Active Appearance Models here are based on the work of Cootes et al
"Active Appearance Models"
https://www.cs.cmu.edu/~efros/courses/LBMV07/Papers/cootes-eccv-98.pdf
Some hints on using the AAMBuilder project:
* You can annotate images with the annotator tool. Note that it is required to
always annotate the images in the same order and always use the same number of points!
* To build an Active Appearance Module one needs a directory having
a set of images with a corresponding points file (same name as the image
but with a different extension). The points files may be created by using
the AAMAnnotation tool.
* The predefined values for building an AAM are quite ok - note that the
model size (aka the models maximum internal image size) has quite some inpact on
the performance.
* You can choose between two different image warping techniques: Linear where the
image is splitted into triangles and these are warped to a different coordinate system or
Thin Plate Splines where the warping is defined by a global spline.
* The library supports a pyramid approach going from the lowest to the highest resolution.
* You can load previously created models.
* To test the model goto the "Test Model Fitting" tab and load an image. Click on the
image somewhere this initializes the fitting algorithm aka the image is placed at the clicked point and
the models tries to fit the underlying image. Examine the result on the right image (the fitting
can take quite some time according to the size of the loaded image, the number of iterations needed and
the warping used).