libidx
Public Member Functions | Public Attributes | Protected Attributes
ebl::gaussian_pyramid< T > Class Template Reference

#include <pyramids.h>

List of all members.

Public Member Functions

 gaussian_pyramid (double c=0.375)
virtual ~gaussian_pyramid ()
 Destructor.
idx< T > reduce (idx< T > &in, uint n=1)
rect< uint > reduce_rect (const rect< uint > &r, uint n=1)
 returns a reduced rectangle by n reductions.
uint count_reductions (uint insz, uint outsz, uint &dist)
 return number of reductions necessary for size insz to be <= outsz
uint count_reductions_exact (rect< uint > &inr, rect< uint > &outr, rect< uint > &inr_exact)
idx< T > expand (idx< T > &in, uint n=1)
rect< uint > expand_rect (const rect< uint > &r, uint n=1)
 returns a expanded rectangle by n expansions.
uint count_expansions (uint insz, uint outsz, uint &dist)
 return number of expansions necessary for size insz to be >= outsz
idx< T > cut_pad (idx< T > &in, int nz)

Public Attributes

idx< T > red_tin

Protected Attributes

double a
idx< T > filter

Detailed Description

template<class T>
class ebl::gaussian_pyramid< T >

A gaussian pyramid implementation. Uses a 5x5 kernel as sepcified in Burt and Adelson paper with a=0.375. This is the same setup used in Matlab too. In 1D, f is given as follows:

f = [1/4-a/2 1/4 a 1/4 1/4-a/2]

The scaling in this implementation fixed to 2, so each downsampled image will be almost half the size in each dimension.

One can create a Laplacian Image Pyramid by successively reducing and expanding an image and taking the difference between two.

Constructor & Destructor Documentation

template<class T >
ebl::gaussian_pyramid< T >::gaussian_pyramid ( double  c = 0.375)

The default value of is 0.375, however one can enter diffferent values to change the shape of the gaussian.

Member Function Documentation

template<class T >
uint ebl::gaussian_pyramid< T >::count_reductions_exact ( rect< uint > &  inr,
rect< uint > &  outr,
rect< uint > &  inr_exact 
)

return number of reductions and set size of the input (closest to current one) to exactly downsample to outr (aspect ratio unchanged).

template<class T >
idx< T > ebl::gaussian_pyramid< T >::cut_pad ( idx< T > &  in,
int  nz 
)

copy in into out, cutting the extra padding of size nz on each border, and return out.

template<class T >
idx< T > ebl::gaussian_pyramid< T >::expand ( idx< T > &  in,
uint  n = 1 
)

The expand method for obtaining an upsampled version of a given image from gaussian pyramid. Optionally specify with n how many expansions to perform. Returns the result. <in> is a -idx2- of -double-. <out> is a -idx2- of -double-. <out> can be any size, it will be resized approriately.

template<class T >
idx< T > ebl::gaussian_pyramid< T >::reduce ( idx< T > &  in,
uint  n = 1 
)

The reduce method for obtaining a scaled down version of the given image. Optionally specify with n how many reductions to perform. Returns the result.

<in> is a -idx2- of -double-. <out> is a -idx2- of -double-. <out> can be any size, it will be resized approriately. Size of <in> might not be suitable for convolving with a 5x5 filter, followed by a 2x2 downsampling. Borders might be removed to satisfy convolution requirements. The original imaage will not be modified, a copy will be generated.

The documentation for this class was generated from the following files: