JP6395750B2 - Signal reconstruction for high dynamic range signals - Google Patents

Description

  The present invention relates generally to images. More specifically, one embodiment of the present invention relates to signal reconstruction of an image having a high dynamic range to improve backward compatibility.

  As used herein, the term “dynamic range” (DR) refers to a range of intensities (eg, luminance, luma) (eg, darkest (black) to brightest white (ie, luminance) in the human visual system (HVS). ))) May be related to the ability to perceive. In this sense, DR relates to the intensity of scene-referred. DR may also relate to the ability of the display device to reasonably or approximately draw an intensity range with a particular width. In this sense, DR relates to the intensity of display-referred. It is intended that the term be used interchangeably (eg, without distinction) anywhere in the specification, unless a specific meaning is specifically designated.

  As used herein, the term high dynamic range (HDR) relates to a DR width that spans 14-15 orders of magnitude in the human visual system (HVS). In practice, a DR that allows a human to perceive a wide range of intensity simultaneously can be somewhat broken with respect to HDR. As used herein, the terms enhanced dynamic range (EDR) or visual dynamic range (VDR), either individually or indistinguishably, include the human visual system (HVS) (including eye movements, some degree of light adaptation change over a scene or image). Is related to a DR that can be perceived in a scene or image. As used herein, EDR can be associated with DR spanning 5-6 orders of magnitude. Thus, although somewhat narrower for true scene-referred HDR, EDR represents a wider DR width and may also be referred to as HDR.

  In practice, the image has one or more color components (eg luma Y and chroma Cb and Cr), and each color component is represented with an accuracy of n bits per pixel (eg n = 8). When linear luminance coding is used, an image with n ≦ 8 (for example, a color 24-bit JPEG image) is a standard dynamic range, and an image with n> 8 is an image with an enhanced dynamic range. EDR and HDR images can also be stored and distributed using high-precision (eg, 16-bit) floating point formats such as the OpenEXR file format developed by Industrial Light and Magic.

  When a video stream is given, information on the encoding parameters is typically embedded in the bitstream as metadata. As used herein, the term “metadata” refers to any auxiliary information that is transmitted as part of the encoded bitstream and helps the decoder render the decoded image. Such metadata includes, but is not limited to, color space or gamut information, reference display parameters, auxiliary signal parameters, and the like, as described herein.

Most consumer desktop displays currently support brightness of 200-300 cd / m ² or nits. Most consumer HDTVs range from 300 to 500 nits, but newer models reach 1000 nits (cd / m ² ). Such conventional displays are therefore typical examples of lower dynamic range (LDR) (also called standard dynamic range (SDR)) for HDR and EDR. With the evolution of both capture devices (eg cameras) and HDR displays (eg Dolby Laboratories' PRM-4200 Professional Reference Monitor), as HDR content becomes more prevalent, HDR content is color-graded to a higher dynamic range (eg 1, 000 nits to 5,000 nits or more). In general, but not limiting, the disclosed method relates to any dynamic range higher than SDR. According to the understanding of the present inventors, an improved technique for encoding a high dynamic range image is desired.

  The approaches described in this section are approaches that can be explored, but not necessarily approaches that have been conceived or explored. Thus, unless otherwise stated to the contrary, none of the approaches described in this section should be considered as eligible for prior art just because they are described in this section. Similarly, unless specified to the contrary, based on this section, problems identified with respect to one or more approaches should not be considered recognized in any prior art.

  One embodiment of the invention is illustrated by way of example and not limitation in the figures of the accompanying drawings. In these figures, similar elements are given the same reference numerals.

FIG. 1 shows an example process for a video distribution pipeline. FIG. 2 illustrates an example process for color conversion to the IPT-PQ color space. FIG. 3 shows an example process for signal reconstruction and encoding. FIG. 4 shows an example tone mapping curve for luminance reconstruction between ST 2084 IPT and BT 1866 IPT in one embodiment of the present invention. FIG. 5 illustrates an example system for backward compatible encoding and decoding using color space reconstruction in one embodiment of the present invention. FIG. 6 illustrates an example process flow for generating a color rotation / scaling matrix in one embodiment of the invention. FIG. 7A shows a hue / saturation reconstruction function in one embodiment of the present invention. FIG. 7B shows a hue / saturation reconstruction function in one embodiment of the present invention. FIG. 8 shows an example of hue / saturation reconstruction between the IPT-PQ and the YCbCr-gamma color space in an embodiment of the present invention. FIG. 9 shows an example of the EETF function in one embodiment of the present invention.

  Signal reconstruction and encoding of high dynamic range (HDR) images is described herein. In the following description, for the sake of convenience, numerous details are set forth in order to provide a thorough understanding of the present invention. However, it will be apparent that the invention may be practiced without these details. On the other hand, details of well-known structures and devices are not described in detail so as not to unnecessarily obscure, obscure, or obfuscate the description of the present invention.

Overview The example embodiments described herein relate to reconstruction and encoding of high dynamic range images. In a method for improving backward compatible decoding, in an encoder, a processor accesses an image database;
Calculating a first hue value of an image in the database in a first color space;
Calculating a second hue value of an image in the database in a second color space;
A hue rotation angle is calculated based on minimizing the hue cost function, wherein the hue cost function is a measure of the difference between the first hue value and the rotated second hue value. Based on
A color rotation matrix is generated based on the hue rotation angle.

  In one embodiment, the first color space is a gamma-based YCbCr color space and the second color space is a PQ-based (PQ-based) IPT color space.

  In one embodiment, the color rotation matrix is used to generate a reconstructed color space based on a suitable color space. An image is encoded using the reconstructed color space, and information about the color rotation matrix is transmitted from the encoder to a decoder.

In one embodiment, a method for reconstructing an input image encoded in a reconstructed color space in a decoder, the decoder comprising:
Receives an encoded input image in a reconstructed color space, wherein the reconstructed color space rotates one or more of the old color spaces by rotating the chroma components of the preferred color space. Generated by approximating parameters,
Accessing metadata transmitted from an encoder to the decoder, wherein the metadata is associated with the encoded input image, and the metadata is
A flag indicating whether a color rotation / scaling matrix exists,
When the flag indicates the presence of a color rotation / scaling matrix, a plurality of coefficients for the color rotation / scaling matrix;
Contains
Generating a decoded image in the reconstructed color space by decoding the encoded input image;
Based on the decoded image in the reconstructed color space and the color rotation / scaling matrix, a decoded image in the preferred color space is generated.

In another embodiment, in the encoder, the processor
Receives an input image in a suitable color space;
Accessing a hue rotation function, wherein the hue rotation function is a hue value for one pixel in the input image in the preferred color space, according to a hue cost condition, and one in an old color space. Generate a rotated hue output value that matches the hue value,
Generating a reconstructed image based on the input image and the hue rotation function;
An encoded reconstructed image is generated by encoding the reconstructed image.

In another embodiment, in the decoder, the processor
Accessing the input image encoded in the reconstructed color space;
Accessing metadata associated with the input image, wherein the metadata is a hue rotation function used to convert the input image from a suitable color space to the reconstructed color space; The hue rotation function is an old color space according to a hue cost condition for a hue value of one pixel in the input image in the preferred color space. Generating a rotated hue output value that matches one of the hue values, and generating an output image in the preferred color space based on the input image and the data associated with the hue rotation function .

Example of Video Supply Processing Pipeline FIG. 1 shows a process example of a conventional video distribution pipeline (100), showing various stages from video capture to video content display. A sequence of video frames (102) is captured or generated using an image generation block (105). Video frames are captured digitally (eg, by a digital camera) or generated by a computer (eg, using computer animation), thereby obtaining video data (107). Alternatively, the video frame (102) may be acquired on film by a silver halide camera. Video data (107) may be obtained by converting the film to a digital format. In the production phase (110), the video data (107) is edited to obtain a video production stream (112).

  The video data of the production stream (112) is then provided to the processor of block (115) for post production editing. Post-production editing (115) adjusts or changes the color or brightness of a specific area of the image to improve the image quality or make the image look specific according to the production intention of the video producer. Can include. This is sometimes called “color timing” or “color grading”. At block (115), the final version (117) for distribution of the production is made by performing other edits (eg, scene selection and sequencing, image cropping, addition of computer generated visual special effects, etc.). You may create it. In post-production editing (115), the video image is viewed on a reference display (125).

  After the post-production (115), the video data (117) as the final production is supplied to a decoding / playback device such as a downstream television receiver, set-top box, movie theater or the like. ). In some embodiments, the encoding block (120) includes audio and audio as defined in ATSC, DVB, DVD, Blu-ray and other delivery formats to generate an encoded bitstream (122). You may have a video encoder. At the receiver, the encoded bitstream (122) is decoded by the decoding unit (130) to produce a decoded signal (132) representing the same or close approximation as the signal (117). ) May be generated. The receiver may be attached to a target display (140) that may have completely different characteristics than the reference display (125). In that case, the display mapping signal (137) is generated using the display management block (135), so that the dynamic range of the decoded signal (132) is mapped to the characteristics of the target display (140). Good.

IPT-PQ Color Space In a preferred embodiment, but not limited to, a portion of the processing pipeline, such as encoding (120), decoding (130), and display management (135), where IPT-PQ color You can do it in what you call space. An example of the use of the IPT-PQ color space for display management applications is described in R.C. Atkins et al. “Display Management for High Dynamic Range Video” WIPO publication WO 2014/130343. The full text is incorporated herein by reference. “Development and testing of a color space (ipt) with implied hue uniformity”, F.M. Ebner and M.M. D. Fairchild, Proc. 6 ^th Color Imaging Conference: Color Science, Systems, and Applications, IS & T, Scottsdale, Ariz. The IPT color space described in 8-13 (hereinafter referred to as Ebner literature, which is incorporated herein in its entirety) is a model of color differences between cones in the human visual system. In this sense, it is like the YCbCr or CIE-Lab color space, but in some scientific studies it has been found that it simulates human visual processing better than these spaces. Similar to CIE-Lab, IPT is a space normalized to some reference luminance. In one embodiment, normalization is based on the maximum brightness of the target display (eg, 5,000 nits).

  As used herein, the term “PQ” refers to perceptual quantization. The human visual system responds very nonlinearly to increased light levels. A person's ability to see a stimulus is affected by the brightness of the stimulus, the magnitude of the stimulus, the spatial frequencies that make up the stimulus, and the brightness level to which the eye has adapted up to the moment of watching the stimulus. In a preferred embodiment, the perceptual quantizer function maps a linear input gray level to an output gray level that is matched by a contrast sensitivity threshold in the human visual system. Examples of PQ mapping functions are described in J. S. Miller et al., US Patent Ser. No. 9,077,994 (referred to as the '994 patent), the entire text of which is incorporated herein by reference. A part thereof is adopted in the name of "High Dynamic Range Electro-Optical Transfer Function of Mastering Reference Display" on August 16, 2014 in the SMPTE ST 2084: 2014 standard. For a fixed stimulus size, for each brightness level (ie, stimulus level), select the minimum visible contrast step at that brightness level, depending on the highest sensitivity adaptation level and the highest sensitivity spatial frequency (according to the HVS model) To do. It represents the response curve of a physical cathode ray tube (CRT) device, compared to the conventional gamma curve, which can have a very rough similarity to how the human visual system responds, '994. The PQ curve determined in the patent simulates the true visual response of the human visual system using a relatively simple functional model.

FIG. 2 illustrates in more detail an example process (200) for color conversion to the IPT-PQ color space in one embodiment. As shown in FIG. 2, when an input signal (202) in a first color space (eg, RGB) is given, the color space conversion in the IPT color space (IPT-PQ) corrected perceptually is as follows: Steps may be included. That is,
a) An optional step (210) may normalize the pixel values (eg, 0 to 4095) of the input signal (202) to pixel values having a dynamic range between 0 and 1.
b) If the input signal (202) is gamma encoded or PQ encoded (eg according to BT.1866 or SMPTE ST 2084), as an optional step (215), the electro-optic transfer function (EOTF) of the signal ) (Provided by the signal metadata) may be used to undo or cancel the code value to luminance conversion by the source display. For example, if the input signal is gamma encoded, an inverse gamma function is applied in this step. If the input signal is PQ encoded according to SMPTE ST 2084, this step applies an inverse PQ function. In practice, the linear signal 217 may be generated by performing the normalization step (210) and the inverse nonlinear encoding (215) using a pre-calculated 1-D lookup table (LUT).
c) In step (220), the linear signal 217 is converted from the original color space (eg, RGB, XYZ, etc.) to the LMS color space. For example, if the original signal is RGB, this step can include two steps. That is, color conversion from RGB to XYZ and color conversion from XYZ to LMS. In one embodiment, but not limited to, XYZ to LMS conversion may be given by:

を導入することにより、ＩＰＴ−ＰＱ色空間での全体的な符号化効率はさらに増大され得る。例えば、ｃ＝０．０２について、クロストーク行列に式（１ａ）の３×３行列を乗算することにより、以下が得られる。

同様に、ｃ＝０．０４について、別の実施形態において、クロストーク行列に、元のＸＹＺからＬＭＳへの行列（例えば式（１ａ））を乗算することにより、以下が得られる。すなわち、

ｄ）Ｅｂｎｅｒ文献によれば、従来のＬＭＳからＩＰＴへの色空間変換は、まずＬＭＳデータに非線形のべき関数を適用し、次に線形変換行列を適用することを含む。データをＬＭＳからＩＰＴに変換したのちにＰＱ関数を適用することによりＩＰＴ−ＰＱドメインに入れることも可能であるが、好適な実施形態においては、ステップ（２２５）でのＬＭＳからＩＰＴへの非線形的符号化のための従来のべき関数に替えて、代わりにＬ、ＭおよびＳ成分の各々のＰＱ非線形的符号化を行う。
ｅ）ＬＭＳからＩＰＴへの線形変換（例えばＥｂｎｅｒ文献に定義される）を用いることにより、ステップ（２３０）は、信号２２２のＩＰＴ−ＰＱ色空間への変換を完了する。例えば、一実施形態において、Ｌ’Ｍ’Ｓ’からＩＰＴ−ＰＱへの変換は、以下によって与えられ得る。

別の実施形態において、実験からＩ’成分はＳ’成分に依存せずに導出することが好ましくあり得ることがわかっており、したがって式（２ａ）は、

となる。 In another embodiment, U.S. Provisional Patent Application Ser. Having the name “Encoding and decoding perceptually-quantized video content”, filed September 26, 2014, which is incorporated herein in its entirety. No. 62 / 056,093 (also filed as PCT / US2015 / 051964 dated September 24, 2015), as part of the XYZ to LMS conversion, the following crosstalk matrix:

, The overall coding efficiency in the IPT-PQ color space can be further increased. For example, for c = 0.02, multiplying the crosstalk matrix by the 3 × 3 matrix of equation (1a) yields:

Similarly, for another embodiment, for c = 0.04, multiplying the crosstalk matrix by the original XYZ to LMS matrix (eg, equation (1a)) yields: That is,

d) According to Ebner literature, conventional color space conversion from LMS to IPT involves first applying a non-linear power function to LMS data and then applying a linear transformation matrix. Although it is possible to enter the IPT-PQ domain by applying the PQ function after converting the data from LMS to IPT, in the preferred embodiment, the non-linear LMS to IPT in step (225) Instead of the conventional power function for encoding, PQ nonlinear encoding of each of the L, M, and S components is performed instead.
e) Step (230) completes the conversion of the signal 222 to the IPT-PQ color space by using a linear transformation from LMS to IPT (eg as defined in Ebner literature). For example, in one embodiment, the conversion from L′ M ′S ′ to IPT-PQ may be given by:

In another embodiment, it has been found from experiments that it may be preferable to derive the I ′ component independent of the S ′ component, so that equation (2a) is

It becomes.

IPT-PQ vs. YCbCr-gamma Most of the existing video compression standards such as MPEG-1, MPEG-2, AVC, HEVC have been tested, evaluated, and optimized for gamma-coded images in the YCbCr color space. Yes. However, experimental results show that the IPT-PQ color space can provide a better representation format for high dynamic range images of 10 bits or more per pixel for each color component. If signal coding is performed in a color space (for example, IPT-PQ) that is more suitable for HDR and a signal with a wider color gamut, better overall image quality can be produced, but older decoders (for example, set-top boxes) are appropriate. Decoding and color conversion may not be possible. According to the understanding of the inventors, a new signal reconstruction technique is required to improve backward compatibility so that even devices that do not know the new color space can generate reasonable images (images). is there.

  FIG. 3 illustrates an example process for signal reconstruction and encoding in one embodiment. As shown in FIG. 3, given the input (302), the forward color reconstruction block (305) may apply a color transformation and / or reconstruction function as needed to apply a suitable color space ( For example, a signal (307) in IPT-PQ-r) is generated. It also generates metadata (309) related to reconstruction, which is communicated to later blocks in the encoding pipeline such as encoder (310), decoder (315), and reverse color reconstruction (320). Also good.

  After receiving the encoded signal (315), the decoder applies decoding (315) (such as HEVC decoding) to generate a decoded signal (317). A decoder that knows the preferred HDR-WCG coded color space (eg, IPT-PQ-r) will apply the proper reverse or reverse reconstruction (320) to ensure that the proper color space is A signal (322) in (for example, IPT-PQ) is generated. The signal (322) may then be converted to YCbCr or RGB for additional post-processing, storage, or display.

  Older decoders that are unaware of the preferred HDR-WCG coding space may treat the HDR-WCG space as an older color space (eg, a gamma-coded YCbCr). However, thanks to the forward color reconstruction (305), the output (317) still has reasonable image quality, although no reverse reconstruction or other color transformations are applied to the decoder output (317). Can have.

Color Reconstruction Consider, but not limited to, the IPT-PQ color space. In one embodiment, the skin tone in the IPT-PQ signal is perceptually matched to the skin color in the YCbCr-gamma signal by generating a linear reconstruction matrix (eg, a 3 × 3 matrix). While such color conversion does not affect the performance of most image processing applications in the IPT color space, it greatly improves color reproduction with older equipment. A similar transformation matrix may be generated so as to match other important colors such as foliage and sky instead of or in addition to the skin color. In one embodiment, the reconstruction matrix may be calculated as follows: That is,

  a) Load a skin color database, such as a reflectance spectrum, and convert them to a device independent color space such as XYZ.

）値を算出、および
ｂ．５）ＹＣｂＣｒ−ガンマ信号の彩度値（例えば

）を算出。 b) The skin color database is converted from XYZ to an old color space format (for example, YCbCr, Rec. 709). This step may include the following sub-steps, for example. That is,
b. 1) Convert database to RGB (Rec. 709),
b. 2) Generate a gamma-coded R′G′B ′ signal by applying gamma to the RGB values (eg compliant with BT.1886),
b. 3) Convert the R′G′B ′ signal into a YCbCr-gamma value (eg, according to Rec. 709),
b. 4) Hue of YCbCr-gamma signal (for example,

) Calculating a value; and b. 5) Saturation value of YCbCr-gamma signal (for example,

).

を算出、および
ｃ．４）彩度値（

）を算出。 c) The skin color value in the database is calculated in a suitable color format (for example, IPT-PQ). This step may include the following sub-steps. That is,
c. 1) Conversion from XYZ to LMS,
c. 2) Conversion from LMS to L'M'S 'and I'P'T' by applying PQ (eg ST 2084 compliant)
c. 3) Hue value (for example,

And c. 4) Saturation value (

).

が、色相に関連するコスト関数を表すとする。ここで、Ｈｕｅ_{ＩＰＴ−ＰＱ−ｒ}は、再構成された色（すなわち、ＩＰＴ−ＰＱ−ｒ）の色相を表しており、

と定義できる。ここで、すべての逆ｔａｎ関数は、（−π，π）で算出される。 d) Calculate a rotation matrix for rotating the IPT values so that the skin color in the rotated or reconstructed IPT-PQ (eg, IPT-PQ-r) matches the skin color in YCbCr-gamma. In one embodiment, this step is calculated by optimizing the cost function for the hue and saturation values of the samples in the two color spaces. For example, in one embodiment, the cost function may represent a mean square error (MSE) between an old color space (eg, YCbCr) and a preferred rotated HDR color space (eg, IPT-PQ). . For example,

Denote a cost function related to hue. Here, Hue _IPT-PQ-r represents the hue of the reconstructed color (ie, IPT-PQ-r),

Can be defined. Here, all inverse tan functions are calculated by (−π, π).

In one embodiment, optimization methods known in the art for finding the value of the angle “a” (denoted a ′) that minimizes the cost function according to certain criteria may be applied. For example, the MATLAB function fminunc (fun, x0) may be applied with fun = Cost _H and x0 = 0.1. For a given a ′, the rotation matrix R may be defined as follows:

である。 As an example, based on a sample database, in one embodiment, for a ′ = 71.74 degrees,

It is.

Given LMS2IPTmat (see, eg, equation (2)), which is a matrix from R and the original L′ M ′S ′ to I′P′T ′, into the reconstructed IPT-PQ-r color space May use the new LMS2IPTmat-r matrix defined as: That is,
LMS2IPTmat−r = R ^T * LMS2IPTmat = ((LMS2IPT mat ^T * R)) ^T (7)
Here, ^AT represents a transposed matrix of the matrix A.

ここで、

ｃ）ｂ’が、Ｃｏｓｔ_Sを最適化するｂ値を表すとする。すると、スケーリングベクトル

をクロマ回転行列に適用することで、１つの色回転・スケーリング３×３行列

を形成し得る。 In one embodiment, saturation may be matched in addition to matching the flesh color hue. This may include the following steps: That is,
a) by applying the R to the original IPT-PQ data to generate a chroma values P _R and T _R data color rotation.
b) Define a saturation cost function, such as the MSE between the saturation values in the original color space and the target color space. That is,

here,

c) _Let b ′ represent the b value that optimizes Cost _S. Then the scaling vector

1 color rotation / scaling 3 × 3 matrix by applying to the chroma rotation matrix

Can be formed.

について、式（４）は、

と変形し得、最適なａ’および最適なｂｉ’（ｉ＝１から４）スケーリング係数の両方について、式（３）を解くことができる。 In some embodiments, the hue and saturation cost functions (eg, equations (3) and (8) may be combined into a single hue / saturation cost function and solved for both a ′ and b ′ simultaneously). For example, from Equation (11), in one embodiment:

Equation (4) is

Equation (3) can be solved for both optimal a ′ and optimal bi ′ (i = 1 to 4) scaling factors.

For example, in one embodiment, for a ′ = 65 degrees and b1 ′ = 1.4, b2 ′ = 1.0, b3 ′ = 1.4, and b4 ′ = 1.0, from Equation (12): can get.

Tone reconstruction The proposed rotation matrix R can improve color reproduction, but the decoded image (317) is still due to differences between nonlinear EOTF encoding functions (eg ST 2084 and BT 1866). May be perceived as having a low contrast. In one embodiment, contrast may be improved by applying a 1-D tone mapping curve to the luminance channel (eg, I ′). This step may include the following sub-steps. That is,
a) Mapping from original content with original HDR maximum brightness (eg 4,000 nits) to SDR target brightness (eg 100 nits) by applying a tone mapping curve (eg sigmoid). An example of such an S-shaped function is described in A. Ballestad and A.M. Kostin, U.S. Patent No. 8,593,480, "Method and Apparatus for Image Data Transformation". An example of another reconstruction function is also disclosed in WIPO publication WO 2014/160705 “Encoding perceptually-quantized video content in multi-layer VDR coding”, which is incorporated herein in its entirety. As I ′ _T = f (I ′) represents the output of the tone mapping function f (),
b) generate linear I _T data by linearizing I ′ _T (eg, applying inverse PQ or gamma function), and c) old-style EOTF encoding (eg, BT) on the linearized I _T signal .1866) generating a gamma encoded luminance signal by applying encoding. This is compressed and sent to the encoder.

  An example of such a mapping between ST 2084 (PQ) and BT 1866 is shown in FIG. This curve has higher midtone contrast, lower blacks, and brighter (less contrast) highlights. This matches the tone scale closer to a standard SDR and allows the image to be viewed even if the input is decoded by an older device. In FIG. 4, although not limited, the input value and the output value are normalized to (0, 1).

  The reconstruction information can be conveyed as metadata from the encoder to the rest of the pipeline. The reconstruction parameters may be determined in various temporal relationships, such as per frame, per scene, or per sequence, to obtain the best possible performance for a given video sequence.

  In this description, the IPT-PQ color space is described, but these methods are equally applicable to other color spaces and color formats. For example, backward compatibility can be improved between different versions of YCbCr (for example, Rec.709 YCbCr and Rec.2020 YCbCr) by applying a similar method. Thus, in one embodiment, using the signal reconstruction method described herein, Rec. By adjusting the 2020 bitstream, the old Rec. Even when decoding is performed using the 709 decoder, a better hue / saturation output can be obtained.

  FIG. 6 illustrates an example process flow for generating a color rotation and scaling matrix in one embodiment. Given an image database (605), step (610) calculates hue and saturation values for the images in the database in a first (old) color space (eg, YCbCr-gamma). Step (615) calculates a hue for an image in the database in a second (preferred) color space (eg, IPT-PQ).

  Given a cost function related to hue (eg, equation (3)), step (620) determines a rotation angle a ′ that is optimal according to some minimized cost condition (eg, mean square error (MSE)). This rotation angle a 'minimizes the distance between the hue group calculated in the old color space and the hue group calculated in the rotated preferred color space. In step (625), a color rotation matrix is generated using the value of a '.

  In addition, a saturation scaler may be calculated as an option. Given a saturation cost function (e.g., Equation 8), step (630) may optionally be performed according to certain minimization cost conditions (e.g., the saturation and color rotation of the signal in the first color space). And the optimal scaler b ′ is determined (640, 645).

  Finally, in step (635), a color rotation / scaling matrix (for example, Equation (11)) is generated by combining the rotation angle and the scaler.

  In the encoder, a color rotation / scaling matrix is applied to input data in a suitable color space to generate data in the reconstructed color space. The data is encoded (compressed) and sent to the decoder along with information related to the color rotation / scaling matrix.

  At the decoder, in the case of an older decoder, the data will be decoded as if it were encoded in an older color space. Despite using the wrong color space information, the dynamic range is reduced, but the image is still viewable with sufficient quality. A newer full-function decoder uses the received metadata information (color rotation / scaling matrix) to decode the image data in a suitable color space, thereby giving the viewer the full high dynamic range of the data. Can be provided to.

SEI message syntax for reconstruction information As described above, in one embodiment, the rotation (R) matrix and scaling vector (S) are changed from L′ M ′S ′ to I′P′T ′ in (230). May be absorbed by the transformation matrix. The tone reconstruction curve may be part of the forward color reconstruction (305). In both cases, adaptive reconstruction information (ie, matrix and tone mapping curves) is obtained from US Provisional Application Ser. No. It can be sent from the encoder to the decoder using the syntax proposed in 62/193390. The full text is incorporated herein by reference.

  In another embodiment, a new color rotation and scale block (510) may be added to the encoder (500A), as shown in FIG. This block may be added after color conversion (200) (eg RGB to IPT-PQ), but is preferably added before forward reconstruction (305). In the decoder (500B), a corresponding inverse color rotation and scaling block (515) may be added after the reverse reconstruction box (320). As shown in FIG. 5, an optional color format conversion box (eg, 4: 4: 4 to 4: 2: 0 (505), or 4: 2: 0 to 4: 4: 4 (520)) is encoded. May be added to the encryption and / or decoding pipeline as needed.

In terms of syntax, a 3 × 3 rotation matrix or simply a 2 × 2 matrix may be specified. This is because typically the luminance channel (eg, Y or I) remains unchanged. Table 1 provides an example of SEI messaging for conveying the color rotation and scaling matrix. However, signal transmission is not limited to SEI messages. It can be inserted into any high level syntax such as SPS, PPS, etc.
Table 1: Example SEI messaging for color rotation and scaling matrix

  A color_rotation_scale_matrix_present_flag equal to 1 indicates that there is a syntax element color_rotation_scale_coeffs [c] [i] for c and i in the range of 0 to 1 (inclusive). The color_rotation_scale_matrix_present_flag equal to 0 indicates that there is no syntax element color_rotation_scale_coeffs [c] [i] for c and i in the range of 0 to 1 (inclusive).

  color_rotation_scale_coeffs [c] [i] specifies the value of 2 × 2 color rotation and scale matrix coefficients. The value of color_rotation_scale_coeffs [c] [i] is in the range of −2 ^ 15 to 2 ^ 15-1 (including both ends). When color_rotation_scale_coeffs [c] [i] is not present, a default color rotation and scale matrix is used.

  In one embodiment, both the encoder and decoder may know the color rotation and scaling matrix (eg, by mutually defining a new color space) and may not need to signal the color rotation matrix from the encoder to the decoder. is there. In another embodiment, the color rotation / scaling matrix may be referenced in a VUI (Video Usability Information) along with IPT-PQ.

が最適なクロマ（色相）回転および彩度スケーリング値を示すとする。ここで、ｈ（ｐ）は画素ｐの色相の尺度を表す。例えば、ＩＰＴ−ＰＱ色空間について、ｆ_Hおよびｆ_S関数は、色相関数ｈ（ｐ）および彩度関数ｓ（ｐ）的に算出することが可能である。すなわち、

Multi-Hue / Saturation Reconstruction In some embodiments, it may be advantageous to apply reconstruction to multiple hues. This increases the accuracy of matching the reconstructed color space to the old color, but incurs additional computational costs at the decoder. For example, consider the problem of optimizing reconstruction for N hues (eg skin color, sky, vegetation, etc.). In one embodiment, by repeating the process described above, the optimal angle and saturation set can be determined as a function of hue. For example, by using a database image for various hue groups, an optimal (rotation angle, saturation scale) value set can be generated. For example, {(a ₁ , b ₁ ), (a ₂ , b ₂ ),. . . , (A _N , b _N )}. Or more generally, for pixel p

Denote the optimal chroma (hue) rotation and saturation scaling values. Here, h (p) represents a measure of the hue of the pixel p. For example, for the IPT-PQ color space, the f _H and f _S functions can be calculated as a color correlation number h (p) and a saturation function s (p). That is,

The functions f _H (h (p)) and f _S (h (p)) may be represented and stored in various ways known in the art, such as look-up tables or piecewise linear or nonlinear polynomials, and It can be transmitted as metadata from the encoder to the decoder.

を適用することにより、適切な再構成された信号を生成する。例えば、ＩＰＴ−ＰＱ色空間について、画素ｐについての再構成されたＰ’およびＴ’色成分は、以下を用いて導出してもよい。すなわち、

Given f _H (h (p)) and f _S (h (p)), the encoder uses the following reconstruction function for each pixel:

To generate an appropriate reconstructed signal. For example, for the IPT-PQ color space, the reconstructed P ′ and T ′ color components for pixel p may be derived using: That is,

を生成する。 At the decoder, the process is reversed. For example, given f _H (h (p)) and f _S (h (p)), from equations (14) and (16), the decoder:

Is generated.

として生成し得る。 To avoid division at the decoder, in some embodiments, the encoder may signal the inverse of f _S (h (p)) (eg, 1 / b (p) value) to the decoder. For the input data in the IPT-PQ space, the original data is

Can be generated as

これらの演算は、コサインおよびサイン関数を算出するために適切なルックアップテーブルを用いることにより、さらに簡素化することができる。 From equation (17), applying inverse reconstruction to recover data in a suitable color space requires trigonometric function operations. In some embodiments, trigonometric functions can be performed using look-up tables. As an example, from the equation (18), the equation (19) can be rewritten as follows.

These operations can be further simplified by using appropriate look-up tables to calculate cosine and sine functions.

  FIG. 7A shows the transformation of the hue from the reconstructed IPT-PQ-r (which appears to older equipment as YCbCr) back to IPT-PQ when the old color space is YCbCr-gamma. An example of the backward reconstruction function is shown. FIG. 7B shows the corresponding backward reconstruction function for adjusting the saturation. FIG. 8 shows how the preferred color space IPT-PQ (820) is adjusted to match the characteristics of the old YCbCr color space (810). Ray (830) represents rotation and scaling.

ここで、ｖ（）およびｗ（）は、再構成された色空間における画像が、旧式の色空間における色相および彩度の組にマッチするように生成された、色相に関連する関数を表す。ｖ（）およびｗ（）関数は、前述のようにメタデータを用いてエンコーダからデコーダへと伝えられてもよく、あるいはエンコーダおよびデコーダの両方が知っている、確立された符号化プロトコルまたは規格の一部であってもよい。 In another embodiment, instead of calculating P and T values as a cosine or sine function of the hue, it is generated based on some other function of the hue (eg, f (tan ⁻¹ (h (p)))). It is also possible to construct a simpler decoder with a lookup table. For example, for a given reconstructed pixel value component P ′ _r (p) and T ′ _r (p), in one embodiment, the decoder may recover the original pixel value as follows: That is,

Here, v () and w () represent hue related functions that are generated so that the image in the reconstructed color space matches the hue and saturation pairs in the old color space. The v () and w () functions may be communicated from the encoder to the decoder using metadata as described above, or of established coding protocols or standards known to both the encoder and decoder. It may be a part.

IC _{TC P} color space IC _{TC P} , also called ICtCp (or IPT), is a new proposed color specifically designed to process high dynamic range and wide color gamut (WCG) signals. It is space. As with ITP-PQ, I (Intensity: intensity) represents the brightness of the PQ encoded signal, the third axis (Tritan Axis) C _T is blue - correspond to the perception of yellow, the first shaft (Protan Axis) C _P corresponds to red-green color perception. In addition to the features of the IPT-PQ mentioned above, in the IC _T C _P,
Rotate chroma and match skin color closer to YCbCr as described above XYZ to LMS matrix optimized for better uniformity and linearity for WCG images L'M'S 'to ICtCp Is optimized to improve the isoluminance and stability for HDR and WCG images.

  As used herein, the term “isoluminous” means a measure of how well the luminance (eg, I of ICtCp or Y ′ of Y′Cb′Cr ′) corresponds to the luminance Y. Indirectly this measures how well a color space separates luma from chroma. From experiments conducted by the inventors, it is known that I of ICtCp corresponds to luma more closely than Y ′ of Y′Cb′Cr ′.

In terms of implementation, the use of IC _T C _P color space requires the same hardware and signal flow and to use conventional gamma encoded YCbCr. For example, consider using YCbCr (Y′Cb′Cr ′) that is gamma corrected in the camera pipeline. Starting from XYZ, the process requires the following steps:
a) Using XYZ to RGB BT. Converted to 2020,
b) Apply inverse EOTF (or OETF) to the output of step a), and c) Apply a 3x3 matrix to the output of step b).

これは式（１ａ）のＸＹＺからＬＭＳへの３×３行列を、ｃ＝０．０４であるクロストーク行列と組み合わせることに相当する（式（１ｃ）も参照のこと）。
ｂ）ステップ（２２５）において、ＰＱ非線形性を適用することにより、前述のように信号（２２２）をＬ’Ｍ’Ｓ’に変換。
ｃ）ステップ（２３０）において、３×３行列を用いてＬ’Ｍ’Ｓ’からＩＣ_TＣ_Pへ変換。この３×３行列は好適な実施形態において、以下のように定義し得る。すなわち、

式（２３）は、式（１２ｂ）の回転行列に、式（２ｂ）の元のＬ’Ｍ’Ｓ’からＩ’Ｐ’Ｔ’への行列を乗算することに相当する。 As shown in FIG. 2, using IC _{TC P} color requires the following steps.
a) In step (220), transform from XYZ to LMS using the following 3 × 3 matrix in the preferred embodiment: That is,

This corresponds to combining the XYZ to LMS 3 × 3 matrix of equation (1a) with a crosstalk matrix where c = 0.04 (see also equation (1c)).
b) In step (225), convert signal (222) to L'M'S 'as described above by applying PQ nonlinearity.
c) In step (230), converted with 3 × 3 matrix from L'M'S 'to IC _T C _P. This 3 × 3 matrix may be defined in the preferred embodiment as follows: That is,

Equation (23) corresponds to multiplying the rotation matrix of Equation (12b) by the matrix from the original L′ M ′S ′ to I′P′T ′ of Equation (2b).

ここで、ＲＧＢ_BT.2020は、ＢＴ．２０２０における３つのＲＧＢ値を表し、ＥＯＴＦ^-1 _ST2084は、ＳＭＰＴＥ ＳＴ ２０８４に従ったＥＯＴＦの逆関数を表す。いくつかの実施形態において、ＥＯＴＦ^-1 _ST2084関数を、ハイブリッドログガンマ（Ｈｙｂｒｉｄ Ｌｏｇ−Ｇａｍｍａ：ＨＬＧ）関数などの、別の非線形的量子化関数により置き換えてもよい。完全な参考が必要であれば、適切な式を表２にも要約しておく。下付き文字Ｄは、ディスプレイ光を意味する。
表２：ＩＣ_TＣ_Pへの色変換

In another embodiment, steps a) to c) can also be expressed as:

Here, RGB _BT.2020 is BT. EOTF- ¹ _ST2084 represents the inverse function of EOTF according to SMPTE ST 2084. In some embodiments, the EOTF ^-1 _ST2084 function may be replaced by another non-linear quantization function, such as a Hybrid Log-Gamma (HLG) function. Appropriate equations are also summarized in Table 2 if full reference is needed. Subscript D means display light.
Table 2: color conversion to the IC _T C _P

を用いて行う。
ｂ）Ｌ’Ｍ’Ｓ’信号から、信号のＥＯＴＦ関数（例えばＳＴ ２０８４の定義に準拠）を用いて、ＬＭＳへと変換。
ｃ）式（２２）の逆行列、例えば

を用いて、ＬＭＳからＸＹＺへと変換。 Conversion back from IC _T C _P to the original color space, follows a similar approach. In one embodiment, this may include the following steps: That is,
a) conversion from IC _T C _P to L'M'S '. This is the inverse matrix of equation (23), ie

To do.
b) Conversion from L'M'S 'signal to LMS using the EOTF function of the signal (e.g. compliant with ST 2084 definition).
c) Inverse matrix of equation (22), eg

To convert from LMS to XYZ.

In one embodiment, the corresponding L′ M ′S ′ to RGB and IC _T C _P to L′ M ′S ′ matrices are given below.

Reference Display Management High dynamic range content may be viewed on a display having a lower dynamic range than the reference display used for mastering that content. In order to view HDR content on a display with a lower dynamic range, it is necessary to perform display mapping. This may take the form of an EETF (electrical-electrical transfer function) in the display. This is typically applied before applying EOTF to the display. This function keeps the balance between preserving intent and maintaining detail by providing a toe and shoulder to elegantly roll off highlights and shadows . FIG. 9 shows an example of EETF mapping from a full dynamic range of 0-10,000 nits to a target display having a capability of 0.1-1,000 nits. This EETF can be introduced into the PQ signal. The plotted points indicate the mapping effect. That is, it shows how the intended light changes to actual display light.

The following are mathematical steps for implementing this tone mapping function for displays with various black and white luminance levels. This EETF, in non-linear domain, IC _T C _P or Y'C _'B C' _R luma channel or to the RGB channels, can be applied individually.
EETF calculation:

  The central region of the tone mapping curve is defined as a one-to-one correspondence from source to target. By calculating additional toe / shoulder roll-off using Hermite splines, the dynamic range is reduced to the capability of the target display.

である。 The spline turning points (toe start (TS) and shoulder start (SS)) are first defined. These are the points where roll-off begins. If minLum and maxLum represent the minimum and maximum brightness values of the target display,

It is.

エルミートスプライン式：

Given E ₁ is the source input signal represented in normalized PQ codeword, the output E ₂ is calculated as follows.

Hermite spline type:

ステップ２：

エルミートスプライン式

ここで

In another embodiment,
Step 1:

Step 2:

Hermite spline type

here

２）Ｙ’Ｃ’_BＣ’_RのＹ’−ＥＥＴＦを介して、Ｙ’Ｃ’_BＣ’_RのルマＹ’チャンネルを処理

Ｙ’₂＝ＥＥＴＦ（Ｙ’₁）
・グレースケールをより正確に調節する
・カラーシフトは限定的
・彩度の変更が必要であり、以下の等式を用いてＣ’_BおよびＣ’_Rチャンネルに適用しなければならない。すなわち、

The resulting EETF curve can be applied to the intensity I channel or Y'C _'B C' luma Y channel _R, the IC _T C _P. Here are a few options to mention:
1) through the I-EETF of IC _T C _P, processing IC _T C _P of the intensity (I) channel

I ₂ = EETF (I ₁ )
• Adjust grayscale more accurately • No color shift • Saturation changes are required and should be applied to the C _T and C _P channels using the following equation: That is,

2) Y'C _'B C' via the Y'-EETF of _R, processing luma Y 'channel Y'C' _B C _'R

Y ′ ₂ = EETF (Y ′ ₁ )
• Adjust grayscale more accurately • Limited color shift • Saturation changes are required and must be applied to the C ′ _B and C ′ _R channels using the following equation: That is,

Embodiments of Computer Systems Embodiments of the present invention include computer systems, systems composed of electronic circuits and components, microcontrollers, field programmable gate arrays (FPGAs) or other configurable or programmable logic devices (PLDs), discrete time Alternatively, it may be implemented using a digital signal processor (DSP), an integrated circuit (IC) device such as an application specific IC (ASIC), and / or an apparatus that includes one or more such systems, devices or components. The computer and / or IC may perform, control, or execute instructions relating to signal reconstruction and encoding of images having an enhanced dynamic range as described herein. The computer and / or IC may compute any of a variety of parameters or values related to the signal reconstruction and encoding process described herein. Image and video embodiments may be implemented in hardware, software, firmware, and various combinations thereof.

  Particular aspects of the present invention include computer processors that execute software instructions to cause a processor to perform the methods of the present invention. For example, one or more processors in a display, encoder, set-top box, transcoder, etc. may execute a software instruction in a program memory accessible to the processor, thereby causing the HDR image signal as described above. Methods for reconstruction and encoding may be implemented. The present invention may be provided in the form of a program product. The program product may include any non-transitory medium that stores a set of computer-readable signals containing instructions for causing the data processor to perform the methods of the invention when executed by the data processor. The program product according to the present invention may take various forms. For example, the program product may include physical media such as floppy disks, magnetic data storage media including hard disk drives, CD ROM, optical data storage media including DVD, electronic data storage media including ROM, flash RAM, and the like. . The computer readable signal on the program product may optionally be compressed or encrypted.

  Where reference is made to a component (eg, a software module, processor, assembly, device, circuit, etc.) above, a reference to that component (including a reference to “means”) is explicitly stated otherwise. Except where noted, any component that performs the function of the component (eg, is functionally equivalent) (not structurally equivalent to the disclosed structure that performs the function that appears in the example embodiments of the present invention described above) As a component equivalent).

Equivalents, extensions, alternatives, etc. Exemplary embodiments for efficient signal reconstruction and coding of HDR images have been described above. In this specification, embodiments of the invention have been described with reference to numerous specific details that may vary from implementation to implementation. Accordingly, the only and exclusive indication as to what the present invention is and what the applicant intends to be is the specific form in which these claims arise, including later corrections. , A set of claims arising from this application. Any definitions expressly set forth herein for terms contained in such claims shall govern the meaning of such terms as used in the claims. Thus, no limitation, element, property, feature, advantage or attribute that is not expressly recited in a claim should limit the scope of the claim in any way. Accordingly, the specification and drawings are to be regarded as illustrative rather than restrictive.

Claims (15)

A method for improving processing of a high dynamic range image, comprising:
By the processor, RGB BT. Accessing a high dynamic range input image in the 2020 color space ;
Color-converting the high dynamic range input image into a first image in an XYZ color space;
Applying a first 3 × 3 color conversion matrix to the first image to generate a second image in a first color space, the first 3 × 3 color conversion matrix Is

The first 3 × 3 color conversion matrix is generated by multiplying the XYZ to LMS 3 × 3 color conversion matrix by the crosstalk matrix;
Generating a third image by applying a non-linear function to each color component of the second image;
Generating an output image by applying a second 3 × 3 color conversion matrix to the third image, wherein the second 3 × 3 color conversion matrix is:

Is included.   The 3 × 3 color conversion matrix from XYZ to LMS is:

Including
  The crosstalk matrix is

The method of claim 1 comprising:   Color-converting the high dynamic range input image into the first image in the XYZ color space, and further applying the first 3 × 3 color conversion matrix to the first image, The step of generating the second image in the first color space includes the RGB BT. By multiplying the color conversion matrix used for the conversion from the 2020 color space to the XYZ color space by the 3 × 3 color conversion matrix from the XYZ to the LMS and the crosstalk matrix, the RGB BT. Generating a third 3 × 3 color conversion matrix for use in conversion from a 2020 color space to the first color space;
  The third 3 × 3 color conversion matrix is

The method of claim 2 comprising: The method of claim 1 , wherein the non-linear function comprises an inverse function of an electro-optic transfer function. 5. The method of claim 4 , wherein the electro-optic transfer function is determined according to the SMPTE ST 2084 standard. A method for reconstructing an input image encoded in a reconstructed color space at a receiver, the reconstructed color space being generated based on minimizing a hue cost function The hue cost function is generated based on a wide gamut space using a rotation matrix, and the hue cost function is a measure of a first hue value in the old color space and a rotated second hue value in the wide gamut space. Based on the difference of and including the following:
And receiving an input image in the reconstructed color space,
Applying a first 3 × 3 color conversion matrix to generate a first image in a first color space, wherein the first 3 × 3 color conversion matrix is:

Including
Generating a second image by applying a non-linear function to each color component of the first image;
By applying a second 3 × 3 color conversion matrix to the second image, RGB BT. Generating an output image in 2020 color space, wherein the second 3 × 3 color transformation matrix is:

Is included. The method of claim 6 , wherein the non-linear function comprises an electro-optic transfer function (EOTF). 8. The method of claim 7 , wherein the electro-optic transfer function is determined according to the SMPTE ST 2084 standard. The method of claim 6 , wherein the non-linear function comprises a hybrid log gamma (HLG) EOTF function. A method for reconstructing an input image encoded in a reconstructed color space at a receiver, the reconstructed color space being generated based on minimizing a hue cost function The hue cost function is generated based on a wide gamut space using a rotation matrix, and the hue cost function is a measure of a first hue value in the old color space and a rotated second hue value in the wide gamut space. Based on the difference of and including the following:
And receiving an input image in the reconstructed color space,
Applying a first 3 × 3 color conversion matrix to generate a first image in a first color space, wherein the first 3 × 3 color conversion matrix is :

Including
Generating a second image by applying a non-linear function to each color component of the first image;
Applying a second 3 × 3 color conversion matrix to the second image to generate a third image in an XYZ color space, wherein the second 3 × 3 color conversion matrix is :

Including
Generating an output image in the target color space by color-converting the third image from an XYZ color space to a target color space; The method of claim 10 , wherein the non-linear function comprises an electro-optic transfer function (EOTF). The method of claim 11 , wherein the electro-optic transfer function is determined according to the SMPTE ST 2084 standard. 11. A method according to claim 6 or claim 10 , wherein the reconstructed color space comprises an ICtCp color space. A processor, the apparatus configured to perform any of the methods according to claims 1 13. A computer-readable non-transitory storage medium storing computer-executable instructions for performing the method according to claims 1 to 13 using one or more processors.

Images