Single layer high dynamic range coding with standard dynamic range backward compatibility

1. A single layer high dynamic range (HDR) coding method with standard dynamic range (SDR) backwards compatibility, the single layer HDR coding method comprising:

generating a tone map using quantized input pixel values, the tone map describing a transformation from data representing a first color component of an HDR video image to data representing a corresponding first color component of an SDR video image on a pixel-by-pixel basis,

modeling the tone map as a piece-wise polynomial reshaping transfer function;

processing one or more additional color components of the HDR video data using the reshaping transfer function to generate SDR video data;

determining an inverse reshaping transfer function from the reshaping transfer function;

generating a self-referential metadata structure describing the inverse reshaping transfer function, the self-referential metadata structure including:

a piece-wise polynomial function describing a mapping of reconstructed SDR samples to HDR samples for the first color components; and

a piece-wise polynomial function describing a mapping of reconstructed SDR samples to HDR samples for the one or more additional color components; and

encoding the SDR video to generate encoded video data including the metadata structure included in a bitstream as supplemental enhancement information data or video usability information data;

wherein the generating of the sell referential data structure describing the inverse reshaping transfer function includes:

dividing the inverse reshaping transfer function into P-1 segments defined by a P pivot points, where P is an integer;

fitting an equation to each segment, each equation having an order, N, where N is an integer that varies based on characteristics of the segment;

for each pivot point except the last pivot point:

allocating a cell in the metadata structure to hold a value indicating a number of coefficients to be stored for the pivot point;

allocating a number of cells in the metadata structure sufficient to describe the coefficients for the pivot point;

storing a value indicating the number of cells for the pivot point into a first one of the allocated cells;

storing the coefficient values in the remaining allocated cells;

further including, far each pivot point except the last pivot point:

allocating a cell in the metadata structure for storing a smoothness degree, the smoothness degree indicating constraints on the equations fitted to the segment, wherein, for each pivot point except the last pivot point:

the number of cells allocated to describe the coefficients of the pivot point is N when the smoothness degree indicates that the functions modeling the segments are continuous; and

the number of cells allocated to describe the coefficients of the pivot point is N-1 when the smoothness degree indicates that the functions modeling the segments are continuous and have corresponding slopes at the pivot points.

2. The method of claim 1 , wherein the number of coefficients to be stored for each of the pivots point except the last pivot point is N+1 where N is the order of the equation for the pivot point.

3. The method of claim 1 , wherein the generating of the self-referential data structure describing the inverse reshaping transfer function includes:

selecting a predetermined transfer function from a plurality of predetermined transfer functions, each of the plurality of predetermined transfer functions having a respectively different number of parameters;

allocating a cell in the metadata structure and storing a value identifying the selected predetermined transfer function in the allocated cell;

adjusting the parameters of the selected predetermined transfer function to approximate the reshaping transfer function;

allocating a number of cells in the metadata structure corresponding to the number of parameters for the selected predetermined transfer function and storing the adjusted parameters for the selected predetermined transfer function in the allocated cells.

4. The method of claim 1 further comprising: receiving linear RGB HDR video data;

applying a coding transfer function to the RGB HDR data to generate non-linear R′G′B′ HDR video data that conforms to the human visual system;

applying a linear color conversion operation to the non-linear R′G′B′ HDR video data to generate color converted video data in a YCbCr color space;

determining a value of a for each Y pixel in the color converted video data, where the transfer function corresponds to pixel-based scaling computed in either the RGB or YCbCr color space and α is a scaling factor of the pixel-based scaling;

wherein:

the generating of the tone map includes:

generating a portion of the tone map for the Y video data;

based on the generated portion of the tone map for the Y video data generating at least one function of the reshaping transfer function for the Cb and Cr video data;

the determining of the inverse reshaping transfer function includes:

generating data describing a first portion of the inverse reshaping transfer function from the tone map for the Y video data;

generating data describing the at least one function to be applied to the Cb and Cr video data; and

the generating of the self-referential metadata structure describing the inverse transfer function includes:

generating a first portion of the metadata describing the first portion of the inverse reshaping transfer function; and

generating a second portion of the metadata describing the at least one function to be applied to the Cb and Cr data.

5. The method of claim 4 wherein:

the data describing the at least one function to be applied to the Cb and Cr video data includes respective Cb and Cr reshaping transfer functions; and

the generating of the second portion of the metadata includes generating metadata describing the respective Cb and Cr reshaping transfer functions.

6. The method of claim 4 wherein:

the data describing the at least one function to be applied to the Cb and Cr video data includes respective scale and offset values for each of the Cb and Cr video data sets, the respective scale and offset values representing modifications of a multiplication of the respective Cb and Cr HDR pixel values by the reshaping transfer function for the Y video data to implement a cross-channel reshaping transform function; and

the generating of the second portion of the metadata includes generating metadata describing the respective scale and offset values for each of the Cb and Cr video data sets.

7. The method of claim 4 , wherein:

the data describing the at least one function to be applied to the Cb and Cr video data includes an input offset value and an input scale value for each of the Cb and Cr video data sets and an output offset value and an output scale value for each of the Cb and Cr video data sets; and

the generating of the second portion of the metadata includes generating metadata describing the respective input and output scale and offset values for each of the Cb and Cr video data sets.

8. The method of claim 4 , further comprising, after applying the linear color conversion operation:

quantizing the color converted video data to a bit depth consistent with the SDR video data; and

downsampling the CbCr video data to generate color converted video data in a 4:2:0 format.

9. Apparatus comprising:

a memory; and

a preprocessor, coupled to the memory, the preprocessor including program instructions that cause the processor to perform single layer high dynamic range (HDR) coding process with standard dynamic range (SDR) backwards compatibility, including instructions to:

generate a tone map using quantized input pixel values, the tone map describing a transformation from data representing a first color component of an HDR video image to data representing a corresponding first color component of an SDR video image on a pixel-by-pixel basis,

model the tone map as a reshaping transfer function;

apply the reshaping transfer function to the HDR to generate SDR video data; determine an inverse reshaping transfer function from the reshaping transfer function;

generate a self-referential metadata structure describing the inverse reshaping transfer function, the self-referential metadata structure including:

a piece-wise polynomial function describing a mapping of reconstructed SDR samples to HDR samples for the first color components; and

a piece-wise polynomial function describing a mapping of reconstructed SDR samples to HDR samples for the one or more additional color components; and

store the self-referential metadata structure in the memory; and an encoder that encodes the SDR video and the self-referential metadata structure from the memory to generate encoded video data including the metadata structure included in a bitstream as supplemental enhancement information data or video usability information data;

wherein the program instructions that cause the preprocessor to generate the self-referential data structure describing the inverse reshaping transfer function include program instructions that cause the preprocessor to:

divide the inverse reshaping transfer function into P-1 segments defined by a P pivot points, where P is an integer;

fit an equation to each segment, each equation having an order, N, where N is an integer that varies based on characteristics of the segment;

for each pivot point except the last pivot point:

allocate a cell in the metadata structure to hold a value indicating a number of coefficients to be stored for the pivot point;

allocate a number of cells in the metadata structure sufficient to describe the coefficients for the pivot point;

store a value indicating the number of cells for the pivot point into a first one of the allocated cells; and

store the coefficient values in the remaining allocated cells;

further including program instructions that, for each pivot point except the last pivot point, cause the preprocessor to:

allocate a cell in the metadata structure for storing a smoothness degree, the smoothness degree indicating constraints on the equations fitted to the segment, wherein, for each pivot point except the last pivot point:

the number of cells allocated to describe the coefficients of the pivot point is N when the smoothness degree indicates that the functions modeling the segments are continuous; and

the number of cells allocated to describe the coefficients of the pivot point is N-1 when the smoothness degree indicates that the functions modeling the segments are continuous and have corresponding slopes at the pivot points.

10. The apparatus of claim 9 , wherein the program instructions that cause the preprocessor to generate the self-referential data structure describing the inverse reshaping transfer function include program instructions that cause the preprocessor to:

select a predetermined transfer function from a plurality of predetermined transfer functions, each of the plurality of predetermined transfer functions having a respectively different number of parameters;

allocate a cell in the metadata structure and store a value identifying the selected predetermined transfer function in the allocated cell;

adjust the parameters of the selected predetermined transfer function to approximate the reshaping transfer function;

allocate a number of cells in the metadata structure corresponding to the number of parameters for the selected predetermined transfer function and storing the adjusted parameters for the selected predetermined transfer function in the allocated cells.

11. The apparatus of claim 9 wherein:

the preprocessor is configured to receive linear RGB HDR video data;

and the program instructions further cause the preprocessor to:

apply a coding transfer function to the RGB HDR data to generate non-linear R′G′B′ HDR video data that conforms to the human visual system;

apply a linear color conversion operation to the non-linear R′G′B′ HDR video data to generate color converted video data in a YCbCr color space;

determining a value of α for each Y pixel in the color converted video data, where the transfer function corresponds to pixel-based scaling computed in either the RGB or YCbCr color spaces and α is a scaling factor of the pixel-based scaling;

wherein:

the program instructions that cause the preprocessor to generate the tone map include instructions that cause the preprocessor to:

generate a portion of the tone map for the Y video data;

based on the generated portion of the tone map for the Y video data generate at least one function of the reshaping transfer function for the Cb and Cr video data;

the program instructions that cause the preprocessor to determine the inverse reshaping transfer function include instructions that cause the preprocessor to:

generate data describing a first portion of the inverse reshaping transfer function from the tone map for the Y video data;

generate data describing at least one function to be applied to the Cb and Cr video data; and

the program instructions that cause the preprocessor to generate the self-referential metadata structure describing the inverse transfer function include program instructions that cause the preprocessor to:

generate a first portion of the metadata describing the first portion of the inverse reshaping transfer function; and

generate a second portion of the metadata describing the at least one function to be applied to the Cb and Cr data.

12. The apparatus of claim 11 wherein:

the data describing the at least one function to be applied to the Cb and Cr video data includes respective Cb and Cr reshaping transfer functions; and

the program instructions that cause the preprocessor to generate the second portion of the metadata include program instructions that cause the preprocessor to generate metadata describing the respective Cb and Cr reshaping transfer functions.

13. The apparatus of claim 11 wherein:

the data describing the at least one function to be applied to the Cb and Cr video data includes respective scale and offset values for each of the Cb and Cr video data sets, the respective scale and offset values representing modifications of a multiplication of the respective Cb and Cr HDR pixel values by the reshaping transfer function for the Y video data to implement a cross-channel reshaping transform function; and

the program instructions cause the preprocessor to generate the second portion of the metadata include program instructions that cause the preprocessor to generate metadata describing the respective scale and offset values for each of the Cb and Cr video data sets.

14. The method of claim 11 , wherein:

the data describing the at least one function to be applied to the Cb and Cr video data includes an input offset value and an input scale value for each of the Cb and Cr video data sets and an output offset value and an output scale value for each of the Cb and Cr video data sets and

the program instructions that cause the preprocessor to generate the second portion of the metadata include program instructions that cause the preprocessor to generate metadata describing the respective input and output scale and offset values for each of the Cb and Cr video data sets.

15. The apparatus of claim 11 , wherein the preprocessor further includes:

a quantizer configured to quantize the color converted video data to a bit depth consistent with the SDR video data; and

a downsampler configured to downsample the CbCr video data to generate color converted video data in a 4:2:0 format.