Active
Shape Model (ASM)
Active Shape Model (ASM) is a model-based methods,
which makes use of a prior model of what is expected in the image, and
typically attempt to find the best match position between the model and the
data in a new image. One can make measurements
whether the target is actually presented after
matching the model. In face recognition application, we collect training
images, represent all shapes with a set of landmarks, to form a Point
Distribution Model (PDM) respectively.
Input face
Landmarks
Landmarks (connected)
Labeled face
After
landmarks alignment and Principal Component Analysis, we construct
gray-level profile for each landmark in all multi-resolution versions of a
training image.
In search procedure, we give the model’s position an initial estimate.
Then it can compute the suggested movements through an iteration
approach using the gray-level profile. When convergence is established,
we get a final matching result.
Example
of image search
According to some disadvantages of ASM,
we adopt a lot of improvements, such as increasing the width
of search profile to reduce the effect of noise, grouping
landmarks to avoid mouth shape distort in the search
procedure and alter the direction of search
profile.
Search result before and after using landmarks grouping
