Ask Your Question
1

LBPH parameters explanation

asked 2017-03-09 07:45:38 -0500

Kelvin gravatar image

Hello, I am testing the LBPH face recognition method varying its parameters, but I don't quite understand the "effect" of each parameter.

The documentation says:

  • radius – The radius used for building the Circular Local Binary Pattern. The greater the radius, the

  • neighbors – The number of sample points to build a Circular Local Binary Pattern from. An appropriate value is to use 8 sample points. Keep in mind: the more sample points you include, the higher the computational cost.

  • grid_x – The number of cells in the horizontal direction, 8 is a common value used in publications. The more cells, the finer the grid, the higher the dimensionality of the resulting feature vector.

  • grid_y – The number of cells in the vertical direction, 8 is a common value used in publications. The more cells, the finer the grid, the higher the dimensionality of the resulting feature vector.

  1. The documentation says: "The greater the radius, the". What happens when I use a greater radius?

  2. As the documentation says in the neighbor's parameter: "the more sample points you include, the higher the computational cost", but what is the benefit of a greater "sample points"?

  3. What is the benefit of using higher grid_x and grid_y? It will increase the computational cost, right? The documentation says: "The more cells, the finer the grid, the higher the dimensionality of the resulting feature vector.". It means that the vector will be more accurate to represent that face?

edit retag flag offensive close merge delete

1 answer

Sort by » oldest newest most voted
3

answered 2017-03-09 08:03:44 -0500

berak gravatar image
  1. not much. if you increase the radius, the sampled region will get larger, but you might miss a fine edge, because the sampled points are further apart.

  2. each additional neighbour will increase the feature vec by a factor of 2 (2 ^ x bins needed per histogram)

  3. if you increase the grid count (and make the patches smaller), less pixels will be in each histogram, making them more "sparse"

in general, larger feature vectors give more precision (up to some point, where you reach the "curse of dimensionality"), but compared to what ? with the defaults, you should get 8 x 8 x 256 = 16384 numbers.

though you're highly encouraged to play with the params there, they seem to be quite optimal already.

edit flag offensive delete link more

Comments

Thanks a lot, @berak, you're always giving the best answers.

Kelvin gravatar imageKelvin ( 2017-03-10 12:25:03 -0500 )edit
Login/Signup to Answer

Question Tools

1 follower

Stats

Asked: 2017-03-09 07:45:38 -0500

Seen: 500 times

Last updated: Mar 09 '17