[FFmpeg-devel] [PATCH v2 5/6] avutil/csp: add EOTF function definitions
Niklas Haas
ffmpeg at haasn.xyz
Sat Nov 30 16:27:53 EET 2024
On Sat, 30 Nov 2024 15:23:50 +0100 Niklas Haas <ffmpeg at haasn.xyz> wrote:
> From: Niklas Haas <git at haasn.dev>
>
> The existing av_csp_trc_func_from_id() mostly implements the OETF, except for
> PQ. As such, we are currently missing a precise definition of an ITU-R EOTF.
> Introduce the new functions av_csp_itu_eotf() and av_csp_itu_eotf_inv(), to fill
> this void. Note that this is not possible in all cases, e.g. AVCOL_TRC_LOG which
> has no corresponding EOTF definition in any ITU-R standard.
>
> Note that we cannot implement the proper HLG and SMPTE 428 OOTFs without access
> to all three color channels, because they are not independent per channel. As a
> result, we need to define a new type to encapsulate a color triplet. I have
> tentatively named this AVColor, in the hopes that it will become useful in other
> contexts as well. The choice of an array rather than explicitly named fields is
> to greatly simplify per-channel processing, matrix multiplications, and the
> like.
Forgot to update the commit message. I decided to nuke AVColor in favor of
passing a double[3] directly. It only makes the code inside the test framework
marginally less pretty, and actually improves the usability of the API in other
contexts IMO.
> ---
> doc/APIchanges | 4 +
> libavutil/csp.c | 205 ++++++++++++++++++++++++++++++++++++++++++++
> libavutil/csp.h | 33 +++++++
> libavutil/version.h | 2 +-
> 4 files changed, 243 insertions(+), 1 deletion(-)
>
> diff --git a/doc/APIchanges b/doc/APIchanges
> index db72121a61..12a2165837 100644
> --- a/doc/APIchanges
> +++ b/doc/APIchanges
> @@ -2,6 +2,10 @@ The last version increases of all libraries were on 2024-03-07
>
> API changes, most recent first:
>
> +2024-11-25 - xxxxxxxxxx - lavu 59.49.100 - csp.h
> + Add struct AVColor, type av_csp_eotf_function, av_csp_itu_eotf() and
> + av_csp_itu_eotf_inv().
> +
> 2024-11-25 - xxxxxxxxxx - lavu 59.48.100 - csp.h
> Add av_csp_trc_func_inv_from_id().
>
> diff --git a/libavutil/csp.c b/libavutil/csp.c
> index fe45c611dd..d69599f513 100644
> --- a/libavutil/csp.c
> +++ b/libavutil/csp.c
> @@ -429,3 +429,208 @@ av_csp_trc_function av_csp_trc_func_inv_from_id(enum AVColorTransferCharacterist
> return NULL;
> return trc_inv_funcs[trc];
> }
> +
> +static void eotf_linear(const double Lw, const double Lb, double E[3])
> +{
> + for (int i = 0; i < 3; i++)
> + E[i] = (Lw - Lb) * E[i] + Lb;
> +}
> +
> +static void eotf_linear_inv(const double Lw, const double Lb, double L[3])
> +{
> + for (int i = 0; i < 3; i++)
> + L[i] = (L[i] - Lb) / (Lw - Lb);
> +}
> +
> +#define WRAP_SDR_OETF(name) \
> +static void oetf_##name(double L[3]) \
> +{ \
> + for (int i = 0; i < 3; i++) \
> + L[i] = trc_##name(L[i]); \
> +} \
> + \
> +static void oetf_##name##_inv(double E[3]) \
> +{ \
> + for (int i = 0; i < 3; i++) \
> + E[i] = trc_##name##_inv(E[i]); \
> +}
> +
> +WRAP_SDR_OETF(gamma22)
> +WRAP_SDR_OETF(gamma28)
> +WRAP_SDR_OETF(iec61966_2_1)
> +
> +#define WRAP_SDR_EOTF(name) \
> +static void eotf_##name(double Lw, double Lb, double E[3]) \
> +{ \
> + oetf_##name##_inv(E); \
> + eotf_linear(Lw, Lb, E); \
> +} \
> + \
> +static void eotf_##name##_inv(double Lw, double Lb, double L[3]) \
> +{ \
> + eotf_linear_inv(Lw, Lb, L); \
> + oetf_##name(L); \
> +}
> +
> +WRAP_SDR_EOTF(gamma22)
> +WRAP_SDR_EOTF(gamma28)
> +WRAP_SDR_EOTF(iec61966_2_1)
> +
> +static void eotf_bt1886(const double Lw, const double Lb, double E[3])
> +{
> + const double Lw_inv = pow(Lw, 1.0 / 2.4);
> + const double Lb_inv = pow(Lb, 1.0 / 2.4);
> + const double a = pow(Lw_inv - Lb_inv, 2.4);
> + const double b = Lb_inv / (Lw_inv - Lb_inv);
> +
> + for (int i = 0; i < 3; i++)
> + E[i] = (-b > E[i]) ? 0.0 : a * pow(E[i] + b, 2.4);
> +}
> +
> +static void eotf_bt1886_inv(const double Lw, const double Lb, double L[3])
> +{
> + const double Lw_inv = pow(Lw, 1.0 / 2.4);
> + const double Lb_inv = pow(Lb, 1.0 / 2.4);
> + const double a = pow(Lw_inv - Lb_inv, 2.4);
> + const double b = Lb_inv / (Lw_inv - Lb_inv);
> +
> + for (int i = 0; i < 3; i++)
> + L[i] = (0.0 > L[i]) ? 0.0 : pow(L[i] / a, 1.0 / 2.4) - b;
> +}
> +
> +static void eotf_smpte_st2084(const double Lw, const double Lb, double E[3])
> +{
> + for (int i = 0; i < 3; i++)
> + E[i] = trc_smpte_st2084_inv(E[i]);
> +}
> +
> +static void eotf_smpte_st2084_inv(const double Lw, const double Lb, double L[3])
> +{
> + for (int i = 0; i < 3; i++)
> + L[i] = trc_smpte_st2084(L[i]);
> +}
> +
> +/* This implementation assumes an SMPTE RP 431-2 reference projector (DCI) */
> +#define DCI_L 48.00
> +#define DCI_P 52.37
> +#define DCI_X (42.94 / DCI_L)
> +#define DCI_Z (45.82 / DCI_L)
> +
> +static void eotf_smpte_st428_1(const double Lw_Y, const double Lb_Y, double E[3])
> +{
> + const double Lw[3] = { DCI_X * Lw_Y, Lw_Y, DCI_Z * Lw_Y };
> + const double Lb[3] = { DCI_X * Lb_Y, Lb_Y, DCI_Z * Lb_Y };
> +
> + for (int i = 0; i < 3; i++) {
> + E[i] = (0.0 > E[i]) ? 0.0 : pow(E[i], 2.6) * DCI_P / DCI_L;
> + E[i] = E[i] * (Lw[i] - Lb[i]) + Lb[i];
> + }
> +}
> +
> +static void eotf_smpte_st428_1_inv(const double Lw_Y, const double Lb_Y, double L[3])
> +{
> + const double Lw[3] = { DCI_X * Lw_Y, Lw_Y, DCI_Z * Lw_Y };
> + const double Lb[3] = { DCI_X * Lb_Y, Lb_Y, DCI_Z * Lb_Y };
> +
> + for (int i = 0; i < 3; i++) {
> + L[i] = (L[i] - Lb[i]) / (Lw[i] - Lb[i]);
> + L[i] = (0.0 > L[i]) ? 0.0 : pow(L[i] * DCI_L / DCI_P, 1.0 / 2.6);
> + }
> +}
> +
> +static void eotf_arib_std_b67(const double Lw, const double Lb, double E[3])
> +{
> + const double gamma = fmax(1.2 + 0.42 * log10(Lw / 1000.0), 1.0);
> +
> + /**
> + * Note: This equation is technically only accurate if the contrast ratio
> + * Lw:Lb is greater than 12:1; otherwise we would need to use a different,
> + * significantly more complicated solution. Ignore this as a highly
> + * degenerate case, since any real world reference display will have a
> + * static contrast ratio multiple orders of magnitude higher.
> + */
> + const double beta = sqrt(3 * pow(Lb / Lw, 1.0 / gamma));
> + double luma;
> +
> + for (int i = 0; i < 3; i++)
> + E[i] = trc_arib_std_b67_inv((1 - beta) * E[i] + beta);
> +
> + luma = 0.2627 * E[0] + 0.6780 * E[1] + 0.0593 * E[2];
> + luma = pow(fmax(luma, 0.0), gamma - 1.0);
> + for (int i = 0; i < 3; i++)
> + E[i] *= Lw * luma;
> +}
> +
> +static void eotf_arib_std_b67_inv(const double Lw, const double Lb, double L[3])
> +{
> + const double gamma = fmax(1.2 + 0.42 * log10(Lw / 1000.0), 1.0);
> + const double beta = sqrt(3 * pow(Lb / Lw, 1 / gamma));
> + double luma = 0.2627 * L[0] + 0.6780 * L[1] + 0.0593 * L[2];
> +
> + if (luma > 0.0) {
> + luma = pow(luma / Lw, (1 - gamma) / gamma);
> + for (int i = 0; i < 3; i++)
> + L[i] *= luma / Lw;
> + } else {
> + L[0] = L[1] = L[2] = 0.0;
> + }
> +
> + for (int i = 0; i < 3; i++)
> + L[i] = (trc_arib_std_b67(L[i]) - beta) / (1 - beta);
> +}
> +
> +static const av_csp_eotf_function eotf_funcs[AVCOL_TRC_NB] = {
> + [AVCOL_TRC_BT709] = eotf_bt1886,
> + [AVCOL_TRC_GAMMA22] = eotf_gamma22,
> + [AVCOL_TRC_GAMMA28] = eotf_gamma28,
> + [AVCOL_TRC_SMPTE170M] = eotf_bt1886,
> + [AVCOL_TRC_SMPTE240M] = eotf_bt1886,
> + [AVCOL_TRC_LINEAR] = eotf_linear,
> + /* There is no EOTF associated with these logarithmic encodings, since they
> + * are defined purely for transmission of scene referred data. */
> + [AVCOL_TRC_LOG] = NULL,
> + [AVCOL_TRC_LOG_SQRT] = NULL,
> + /* BT.1886 is already defined for values below 0.0, as far as physically
> + * meaningful, so we can directly use it for extended range encodings */
> + [AVCOL_TRC_IEC61966_2_4] = eotf_bt1886,
> + [AVCOL_TRC_BT1361_ECG] = eotf_bt1886,
> + [AVCOL_TRC_IEC61966_2_1] = eotf_iec61966_2_1,
> + [AVCOL_TRC_BT2020_10] = eotf_bt1886,
> + [AVCOL_TRC_BT2020_12] = eotf_bt1886,
> + [AVCOL_TRC_SMPTE2084] = eotf_smpte_st2084,
> + [AVCOL_TRC_SMPTE428] = eotf_smpte_st428_1,
> + [AVCOL_TRC_ARIB_STD_B67] = eotf_arib_std_b67,
> +};
> +
> +av_csp_eotf_function av_csp_itu_eotf(enum AVColorTransferCharacteristic trc)
> +{
> + if (trc < 0 || trc >= AVCOL_TRC_NB)
> + return NULL;
> + return eotf_funcs[trc];
> +}
> +
> +static const av_csp_eotf_function eotf_inv_funcs[AVCOL_TRC_NB] = {
> + [AVCOL_TRC_BT709] = eotf_bt1886_inv,
> + [AVCOL_TRC_GAMMA22] = eotf_gamma22_inv,
> + [AVCOL_TRC_GAMMA28] = eotf_gamma28_inv,
> + [AVCOL_TRC_SMPTE170M] = eotf_bt1886_inv,
> + [AVCOL_TRC_SMPTE240M] = eotf_bt1886_inv,
> + [AVCOL_TRC_LINEAR] = eotf_linear_inv,
> + [AVCOL_TRC_LOG] = NULL,
> + [AVCOL_TRC_LOG_SQRT] = NULL,
> + [AVCOL_TRC_IEC61966_2_4] = eotf_bt1886_inv,
> + [AVCOL_TRC_BT1361_ECG] = eotf_bt1886_inv,
> + [AVCOL_TRC_IEC61966_2_1] = eotf_iec61966_2_1_inv,
> + [AVCOL_TRC_BT2020_10] = eotf_bt1886_inv,
> + [AVCOL_TRC_BT2020_12] = eotf_bt1886_inv,
> + [AVCOL_TRC_SMPTE2084] = eotf_smpte_st2084_inv,
> + [AVCOL_TRC_SMPTE428] = eotf_smpte_st428_1_inv,
> + [AVCOL_TRC_ARIB_STD_B67] = eotf_arib_std_b67_inv,
> +};
> +
> +av_csp_eotf_function av_csp_itu_eotf_inv(enum AVColorTransferCharacteristic trc)
> +{
> + if (trc < 0 || trc >= AVCOL_TRC_NB)
> + return NULL;
> + return eotf_inv_funcs[trc];
> +}
> diff --git a/libavutil/csp.h b/libavutil/csp.h
> index 58deebcb44..9b74c631d2 100644
> --- a/libavutil/csp.h
> +++ b/libavutil/csp.h
> @@ -152,6 +152,39 @@ av_csp_trc_function av_csp_trc_func_from_id(enum AVColorTransferCharacteristic t
> */
> av_csp_trc_function av_csp_trc_func_inv_from_id(enum AVColorTransferCharacteristic trc);
>
> +/**
> + * Function pointer representing an ITU EOTF transfer for a given reference
> + * display configuration.
> + *
> + * @param Lw The white point luminance of the display, in nits (cd/m^2).
> + * @param Lb The black point luminance of the display, in nits (cd/m^2).
> + */
> +typedef void (*av_csp_eotf_function)(double Lw, double Lb, double c[3]);
> +
> +/**
> + * Returns the ITU EOTF corresponding to a given TRC. This converts from the
> + * signal level [0,1] to the raw output display luminance in nits (cd/m^2).
> + * This is done per channel in RGB space, except for AVCOL_TRC_SMPTE428, which
> + * assumes CIE XYZ in- and output.
> + *
> + * @return A pointer to the function implementing the given TRC, or NULL if no
> + * such function is defined.
> + *
> + * @note In general, the resulting function is defined (wherever possible) for
> + * out-of-range values, even though these values do not have a physical
> + * meaning on the given display. Users should clamp inputs (or outputs)
> + * if this behavior is not desired.
> + *
> + * This is also the case for functions like PQ, which are defined over an
> + * absolute signal range independent of the target display capabilities.
> + */
> +av_csp_eotf_function av_csp_itu_eotf(enum AVColorTransferCharacteristic trc);
> +
> +/**
> + * Returns the mathematical inverse of the corresponding EOTF.
> + */
> +av_csp_eotf_function av_csp_itu_eotf_inv(enum AVColorTransferCharacteristic trc);
> +
> /**
> * @}
> */
> diff --git a/libavutil/version.h b/libavutil/version.h
> index efc569e090..ad1f3e68ba 100644
> --- a/libavutil/version.h
> +++ b/libavutil/version.h
> @@ -79,7 +79,7 @@
> */
>
> #define LIBAVUTIL_VERSION_MAJOR 59
> -#define LIBAVUTIL_VERSION_MINOR 48
> +#define LIBAVUTIL_VERSION_MINOR 49
> #define LIBAVUTIL_VERSION_MICRO 100
>
> #define LIBAVUTIL_VERSION_INT AV_VERSION_INT(LIBAVUTIL_VERSION_MAJOR, \
> --
> 2.47.0
>
More information about the ffmpeg-devel
mailing list