Quick Start

Build & Install

WaveletCV is built and installed using cmake 3.24 or newer.

# Download
git clone https://github.com/cindolfi/waveletcv.git
cd waveletcv
git checkout v0.1.0

# Configure
mkdir build
cmake -B build -DCMAKE_BUILD_TYPE=Release

# Build
cmake ---build build

# Install To /usr/local
sudo cmake --install build

Basic Usage

Wavelet Objects

#include <wtcv/wavelet.hpp>

Builtin wavelets are created by name

wtcv::Wavelet wavelet = wtcv::Wavelet::create("db2");

or by factory

wtcv::Wavelet wavelet = wtcv::create_daubuchies(2);

Accessing the filter banks decomposition and reconstruction kernels

wavelet.filter_bank().decompose_kernels().lowpass()
wavelet.filter_bank().decompose_kernels().highpass()
wavelet.filter_bank().reconstruct_kernels().lowpass()
wavelet.filter_bank().reconstruct_kernels().highpass()

See also

Discrete Wavelet Transform (DWT)

#include <wtcv/dwt2d.hpp>

Performing a discrete wavelet transformation (DWT) of an image is done using a functional style

cv::Mat image = cv::imread(filename);
DWT2D::Coeffs coeffs = wtcv::dwt2d(image, "db2");

or an object oriented approach

wtcv::Wavelet wavelet = wtcv::Wavelet::create("db2");
int levels = 2;
wtcv::DWT2D dwt(wavelet);
wtcv::DWT2D::Coeffs coeffs = dwt(image, levels);

Reconstruct the image by inverting the DWT

cv::Mat reconstructed_image = coeffs.reconstruct();

Accessing DWT Coefficients

Note

The horizontal detail coefficients are used for illustration. There are corresponding accessors for vertical and diagonal detail coefficients.

Access the approximation coefficients

cv::Mat approx_coeffs = coeffs.approx();

Access the finest scale (i.e. highest resolution) horizontal subband coefficients

cv::Mat finest_horizontal_coeffs = coeffs.horizontal_detail(0);
coeffs.set_horizontal_detail(0, finest_horizontal_coeffs);

Or use the parameterized subband version

cv::Mat finest_horizontal_coeffs = coeffs.detail(wtcv::HORIZONTAL, 0);
coeffs.set_detail(0, wtcv::HORIZONTAL, finest_horizontal_coeffs);

Negative Level Indexing

Use negative level indexing to access the coarsest scale (i.e. lowest resolution) horizontal subband coefficients

// Equivalent to coeffs.horizontal_detail(coeffs.levels() - 1)
cv::Mat coarsest_horizontal_coeffs = coeffs.horizontal_detail(-1);
coeffs.set_horizontal_detail(-1, coarsest_horizontal_coeffs);

Collect Details At Multiple Scales

Get horizontal detail coefficients at every scale

std::vector<cv::Mat> horizontal_details = coeffs.collect_horizontal_details();

Get detail coefficients at every scale and subband

std::vector<wtcv::DWT2D::Coeffs::DetailTuple>> details = coeffs.details();

See also

Shrink DWT Coefficients

#include <wtcv/shrink.hpp>
wtcv::DWT2D::Coeffs coeffs = ...;

Shrinking DWT coefficients is the basis for many denoising and compression applications. There are several shrinking algorithms implemented. Take the BayesShrink algorithm as an example

coeffs = wtcv::bayes_shrink(coeffs);

Alternatively, the object oriented API can be used in a polymorphic way

wtcv::Shrinker* shrinker = new wtcv::BayesShrinker();
coeffs = shrinker->shrink(coeffs);

or as a function object

wtcv::BayesShrinker shrink;
coeffs = shrink(coeffs);

The functional API is simpler and more succinct, whereas the object oriented API offers more options to fine tune the algorithm.

See also