CODEWITHSUNDEEP
computer-vision
theotherphil
theotherphil

Reputation: 667

What are good filters to use for feature generation with boosting

I'm trying to implement something like the method given here:

research.microsoft.com/en-us/um/people/viola/Pubs/MIT/tieuIJCV.pdf

Images "similar" to a query image are retrieved from a database by using AdaBoost on features generated from the images by repeatedly filtering and downsizing them.

The paper shows images of the filters it uses for detecting basic images features (vertical lines, diagonals, corners, etc.) and states that these are separable but doesn't say exactly what the matrices are for these features. Does anyone know which matrices were used for the filters, or in general what are good choices of filters for detecting simple geometric shapes for use with boosting.

Thanks

Upvotes: 1

Views: 77

Answers (1)

tomfa
tomfa

Reputation: 11

They seem to be similar to Haar features that's described by the same author in his first boosting paper.

click here

I've only used the haar features when implemented the AdaBoost in class. In general I guess you could make your own haar filters, some simple examples can be seen here.

Upvotes: 1

Related Questions