RGBCube/serenity
1
Fork 0
mirror of https://github.com/RGBCube/serenity synced 2025-07-25 17:07:34 +00:00
Code Issues Activity Actions

ICC: Implement TRC inversion in from_pcs for parametric curves

Browse source
This commit is contained in:
Nico Weber 2023-05-01 11:14:10 -04:00 • committed by Sam Atkins
parent 4169c94ebe
commit 9bd35fda56
3 changed files with 71 additions and 8 deletions
Download patch file Download diff file Expand all files Collapse all files

15
Tests/LibGfx/TestICCProfile.cpp
View file

@ -158,6 +158,10 @@ TEST_CASE(from_pcs)
{
auto sRGB = MUST(Gfx::ICC::sRGB());
auto sRGB_curve_pointer = MUST(Gfx::ICC::sRGB_curve());
VERIFY(sRGB_curve_pointer->type() == Gfx::ICC::ParametricCurveTagData::Type);
auto const& sRGB_curve = static_cast<Gfx::ICC::ParametricCurveTagData const&>(*sRGB_curve_pointer);
auto sRGB_from_xyz = [&sRGB](FloatVector3 const& XYZ) {
u8 rgb[3];
MUST(sRGB->from_pcs(XYZ, rgb));
@ -185,7 +189,16 @@ TEST_CASE(from_pcs)
EXPECT_EQ(sRGB_from_xyz(g_xyz + b_xyz), Color(0, 255, 255));
EXPECT_EQ(sRGB_from_xyz(r_xyz + g_xyz + b_xyz), Color(255, 255, 255));
// FIXME: Implement and test the inverse curve transform.
// Test the inverse curve transform.
float f64 = sRGB_curve.evaluate(64 / 255.f);
EXPECT_EQ(sRGB_from_xyz((r_xyz + g_xyz + b_xyz) * f64), Color(64, 64, 64));
float f128 = sRGB_curve.evaluate(128 / 255.f);
EXPECT_EQ(sRGB_from_xyz((r_xyz + g_xyz + b_xyz) * f128), Color(128, 128, 128));
// Test for curve and matrix combined.
float f192 = sRGB_curve.evaluate(192 / 255.f);
EXPECT_EQ(sRGB_from_xyz(r_xyz * f64 + g_xyz * f128 + b_xyz * f192), Color(64, 128, 192));
}
TEST_CASE(to_lab)

25
Userland/Libraries/LibGfx/ICC/Profile.cpp
View file

@ -1551,6 +1551,7 @@ ErrorOr<void> Profile::from_pcs(FloatVector3 const& pcs, Bytes color) const
// greenTRC^-1, and blueTRC^-1 function is undefined. If a one-dimensional curve is constant, the curve cannot be
// inverted."
// Convert from XYZ to linear rgb.
// FIXME: Inverting matrix and curve on every call to this function is very inefficient.
auto const& red_matrix_column = this->red_matrix_column();
auto const& green_matrix_column = this->green_matrix_column();
@ -1561,13 +1562,22 @@ ErrorOr<void> Profile::from_pcs(FloatVector3 const& pcs, Bytes color) const
red_matrix_column.Y, green_matrix_column.Y, blue_matrix_column.Y,
red_matrix_column.Z, green_matrix_column.Z, blue_matrix_column.Z
};
if (!forward_matrix.is_invertible())
return Error::from_string_literal("ICC::Profile::from_pcs: matrix not invertible");
auto matrix = forward_matrix.inverse();
FloatVector3 linear_rgb = matrix * pcs;
auto evaluate_curve_inverse = [this](TagSignature curve_tag, float f) {
auto const& trc = *m_tag_table.get(curve_tag).value();
VERIFY(trc.type() == CurveTagData::Type || trc.type() == ParametricCurveTagData::Type);
if (trc.type() == CurveTagData::Type) {
TODO();
return 0.f;
}
return static_cast<ParametricCurveTagData const&>(trc).evaluate_inverse(f);
};
// Convert from linear rgb to device rgb.
// See equations (F.8) - (F.16) above.
// FIXME: The spec says to do this, but it loses information. Color.js returns unclamped
// values instead (...but how do those make it through the TRC?) and has a separate
@ -1576,12 +1586,13 @@ ErrorOr<void> Profile::from_pcs(FloatVector3 const& pcs, Bytes color) const
// (For LUT profiles, I think the gamut mapping is baked into the BToA* data in the profile (?).
// But for matrix profiles, it'd have to be done in code.)
linear_rgb.clamp(0.f, 1.f);
float device_r = evaluate_curve_inverse(redTRCTag, linear_rgb[0]);
float device_g = evaluate_curve_inverse(greenTRCTag, linear_rgb[1]);
float device_b = evaluate_curve_inverse(blueTRCTag, linear_rgb[2]);
// FIXME: Implement curve inversion and apply inverse curve transform here.
color[0] = round(255 * linear_rgb[0]);
color[1] = round(255 * linear_rgb[1]);
color[2] = round(255 * linear_rgb[2]);
color[0] = round(255 * device_r);
color[1] = round(255 * device_g);
color[2] = round(255 * device_b);
return {};
}

39
Userland/Libraries/LibGfx/ICC/TagTypes.h
View file

@ -739,6 +739,45 @@ public:
VERIFY_NOT_REACHED();
}
// y must be in [0..1].
float evaluate_inverse(float y) const
{
VERIFY(0.f <= y && y <= 1.f);
// See "Recommendations" section in https://www.color.org/whitepapers/ICC_White_Paper35-Use_of_the_parametricCurveType.pdf
// Requirements for the curve to be non-decreasing:
// * γ > 0
// * a > 0 for types 1-4
// * c ≥ 0 for types 3 and 4
//
// Types 3 and 4 additionally require:
// To prevent negative discontinuities:
// * cd ≤ (ad + b) for type 3
// * cd + f ≤ (ad + b)^γ + e for type 4
// To prevent complex numbers:
// * ad + b ≥ 0
// FIXME: Check these requirements somewhere.
switch (function_type()) {
case FunctionType::Type0:
return powf(y, 1.f / (float)g());
case FunctionType::Type1:
return (powf(y, 1.f / (float)g()) - (float)b()) / (float)a();
case FunctionType::Type2:
// Only defined for Y >= c, so I suppose this requires c <= 0 in practice (?).
return (powf(y - (float)c(), 1.f / (float)g()) - (float)b()) / (float)a();
case FunctionType::Type3:
if (y >= (float)c() * (float)d())
return (powf(y, 1.f / (float)g()) - (float)b()) / (float)a();
return y / (float)c();
case FunctionType::Type4:
if (y >= (float)c() * (float)d())
return (powf(y - (float)e(), 1.f / (float)g()) - (float)b()) / (float)a();
return (y - (float)f()) / (float)c();
}
VERIFY_NOT_REACHED();
}
private:
FunctionType m_function_type;