numeric.hpp

//
// Copyright 2019 Olzhas Zhumabek <anonymous.from.applecity@gmail.com>
// Copyright 2021 Pranam Lashkari <plashkari628@gmail.com>
//
// Use, modification and distribution are subject to the Boost Software License,
// Version 1.0. (See accompanying file LICENSE_1_0.txt or copy at
// http://www.boost.org/LICENSE_1_0.txt)
//
#ifndef BOOST_GIL_IMAGE_PROCESSING_NUMERIC_HPP
#define BOOST_GIL_IMAGE_PROCESSING_NUMERIC_HPP
 
#include <boost/gil/image_processing/kernel.hpp>
#include <boost/gil/image_processing/convolve.hpp>
#include <boost/gil/image_view.hpp>
#include <boost/gil/typedefs.hpp>
#include <boost/gil/detail/math.hpp>
// fixes ambiguous call to std::abs, https://stackoverflow.com/a/30084734/4593721
#include <cstdlib>
#include <cmath>
 
namespace boost { namespace gil {
 
inline double normalized_sinc(double x)
{
    return std::sin(x * boost::gil::detail::pi) / (x * boost::gil::detail::pi);
}
 
inline double lanczos(double x, std::ptrdiff_t a)
{
    // means == but <= avoids compiler warning
    if (0 <= x && x <= 0)
        return 1;
 
    if (static_cast<double>(-a) < x && x < static_cast<double>(a))
        return normalized_sinc(x) / normalized_sinc(x / static_cast<double>(a));
 
    return 0;
}
inline double lanczos(double x, std::ptrdiff_t a) {…}
 
#if BOOST_WORKAROUND(BOOST_MSVC, >= 1400)
#pragma warning(push)
#pragma warning(disable:4244) // 'argument': conversion from 'const Channel' to 'BaseChannelValue', possible loss of data
#endif
 
inline void compute_tensor_entries(
    boost::gil::gray16s_view_t dx,
    boost::gil::gray16s_view_t dy,
    boost::gil::gray32f_view_t m11,
    boost::gil::gray32f_view_t m12_21,
    boost::gil::gray32f_view_t m22)
{
    for (std::ptrdiff_t y = 0; y < dx.height(); ++y) {
        for (std::ptrdiff_t x = 0; x < dx.width(); ++x) {
            auto dx_value = dx(x, y);
            auto dy_value = dy(x, y);
            m11(x, y) = dx_value * dx_value;
            m12_21(x, y) = dx_value * dy_value;
            m22(x, y) = dy_value * dy_value;
        }
    }
}
 
#if BOOST_WORKAROUND(BOOST_MSVC, >= 1400)
#pragma warning(pop)
#endif
 
template <typename T = float, typename Allocator = std::allocator<T>>
inline auto generate_normalized_mean(std::size_t side_length)
    -> detail::kernel_2d<T, Allocator>
{
    if (side_length % 2 != 1)
        throw std::invalid_argument("kernel dimensions should be odd and equal");
    const float entry = 1.0f / static_cast<float>(side_length * side_length);
 
    detail::kernel_2d<T, Allocator> result(side_length, side_length / 2, side_length / 2);
    for (auto& cell: result) {
        cell = entry;
    }
 
    return result;
}
inline auto generate_normalized_mean(std::size_t side_length) {…}
 
template <typename T = float, typename Allocator = std::allocator<T>>
inline auto generate_unnormalized_mean(std::size_t side_length)
    -> detail::kernel_2d<T, Allocator>
{
    if (side_length % 2 != 1)
        throw std::invalid_argument("kernel dimensions should be odd and equal");
 
    detail::kernel_2d<T, Allocator> result(side_length, side_length / 2, side_length / 2);
    for (auto& cell: result) {
        cell = 1.0f;
    }
 
    return result;
}
inline auto generate_unnormalized_mean(std::size_t side_length) {…}
 
template <typename T = float, typename Allocator = std::allocator<T>>
inline auto generate_gaussian_kernel(std::size_t side_length, double sigma)
    -> detail::kernel_2d<T, Allocator>
{
    if (side_length % 2 != 1)
        throw std::invalid_argument("kernel dimensions should be odd and equal");
 
    const double denominator = 2 * boost::gil::detail::pi * sigma * sigma;
    auto const middle = side_length / 2;
    std::vector<T, Allocator> values(side_length * side_length);
    T sum{0};
    for (std::size_t y = 0; y < side_length; ++y)
    {
        for (std::size_t x = 0; x < side_length; ++x)
        {
            const auto delta_x = x - middle;
            const auto delta_y = y - middle;
            const auto power = static_cast<double>(delta_x * delta_x + delta_y * delta_y) / (2 * sigma * sigma);
            const double nominator = std::exp(-power);
            const auto value = static_cast<T>(nominator / denominator);
            values[y * side_length + x] = value;
            sum += value;
        }
    }
 
    // normalize so that Gaussian kernel sums up to 1.
    std::transform(values.begin(), values.end(), values.begin(), [&sum](const auto & v) { return v/sum; });
 
    return detail::kernel_2d<T, Allocator>(values.begin(), values.size(), middle, middle);
}
inline auto generate_gaussian_kernel(std::size_t side_length, double sigma) {…}
 
template <typename T = float, typename Allocator = std::allocator<T>>
inline auto generate_dx_sobel(unsigned int degree = 1)
    -> detail::kernel_2d<T, Allocator>
{
    switch (degree)
    {
        case 0:
        {
            return detail::get_identity_kernel<T, Allocator>();
        }
        case 1:
        {
            detail::kernel_2d<T, Allocator> result(3, 1, 1);
            std::copy(detail::dx_sobel.begin(), detail::dx_sobel.end(), result.begin());
            return result;
        }
        default:
            throw std::logic_error("not supported yet");
    }
 
    //to not upset compiler
    throw std::runtime_error("unreachable statement");
}
inline auto generate_dx_sobel(unsigned int degree = 1) {…}
 
template <typename T = float, typename Allocator = std::allocator<T>>
inline auto generate_dx_scharr(unsigned int degree = 1)
    -> detail::kernel_2d<T, Allocator>
{
    switch (degree)
    {
        case 0:
        {
            return detail::get_identity_kernel<T, Allocator>();
        }
        case 1:
        {
            detail::kernel_2d<T, Allocator> result(3, 1, 1);
            std::copy(detail::dx_scharr.begin(), detail::dx_scharr.end(), result.begin());
            return result;
        }
        default:
            throw std::logic_error("not supported yet");
    }
 
    //to not upset compiler
    throw std::runtime_error("unreachable statement");
}
inline auto generate_dx_scharr(unsigned int degree = 1) {…}
 
template <typename T = float, typename Allocator = std::allocator<T>>
inline auto generate_dy_sobel(unsigned int degree = 1)
    -> detail::kernel_2d<T, Allocator>
{
    switch (degree)
    {
        case 0:
        {
            return detail::get_identity_kernel<T, Allocator>();
        }
        case 1:
        {
            detail::kernel_2d<T, Allocator> result(3, 1, 1);
            std::copy(detail::dy_sobel.begin(), detail::dy_sobel.end(), result.begin());
            return result;
        }
        default:
            throw std::logic_error("not supported yet");
    }
 
    //to not upset compiler
    throw std::runtime_error("unreachable statement");
}
inline auto generate_dy_sobel(unsigned int degree = 1) {…}
 
template <typename T = float, typename Allocator = std::allocator<T>>
inline auto generate_dy_scharr(unsigned int degree = 1)
    -> detail::kernel_2d<T, Allocator> 
{
    switch (degree)
    {
        case 0:
        {
            return detail::get_identity_kernel<T, Allocator>();
        }
        case 1:
        {
            detail::kernel_2d<T, Allocator> result(3, 1, 1);
            std::copy(detail::dy_scharr.begin(), detail::dy_scharr.end(), result.begin());
            return result;
        }
        default:
            throw std::logic_error("not supported yet");
    }
 
    //to not upset compiler
    throw std::runtime_error("unreachable statement");
}
inline auto generate_dy_scharr(unsigned int degree = 1) {…}
 
template <typename GradientView, typename OutputView>
inline void compute_hessian_entries(
    GradientView dx,
    GradientView dy,
    OutputView ddxx,
    OutputView dxdy,
    OutputView ddyy)
{
    auto sobel_x = generate_dx_sobel();
    auto sobel_y = generate_dy_sobel();
    detail::convolve_2d(dx, sobel_x, ddxx);
    detail::convolve_2d(dx, sobel_y, dxdy);
    detail::convolve_2d(dy, sobel_y, ddyy);
}
inline void compute_hessian_entries( {…}
 
}} // namespace boost::gil
 
#endif