Automatic display management metadata generation for gaming and/or SDR+ contents
Different candidate image data feature types are evaluated to identify one or more specific image data feature types to be used in training a prediction model for optimizing one or more image metadata parameters. A plurality of image data features of the one or more selected image data feature types is extracted from one or more images. The plurality of image data features of the one or more selected image data feature types is reduced into a plurality of significant image data features. A total number of image data features in the plurality of significant image data features is no larger than a total number of image data features in the plurality of image data features of the one or more selected image data feature types. The plurality of significant image data features is applied to training the prediction model for optimizing one or more image metadata parameters.
1. A method for image metadata optimization, the method comprising:
determining a plurality of sets of training images and a plurality of sets of targeted values for one or more image metadata parameters, each set of targeted values in the plurality of sets of targeted values corresponding to a respective set of one or more training images in the plurality of sets of training images;
extracting a plurality of sets of image data features of one or more selected image data feature types from the plurality of sets of training images, each set of image data features in the plurality of sets of image data features corresponding to a respective set of one or more training images in the plurality of sets of training images;
applying the plurality of sets of image data features of the one or more selected image data feature types and the plurality of sets of targeted values for the one or more image metadata parameters to training one or more prediction models for optimizing the one or more image metadata parameters;
using the one or more prediction models to generate one or more optimized values for the one or more image metadata parameters for one or more non-training images;
causing the one or more optimized values for the one or more image metadata parameters to be used in rendering the one or more non-training images;
the method being characterized in that the one or more non-training images represent one or more standard dynamic range (SDR) images; wherein the plurality of sets of training images represent high dynamic range (HDR) images; the method further being characterized by it comprising:
extracting a set of non-training SDR image data features of the one or more selected image data feature types from the one or more non-training images;
generating a set of non-training HDR image data features of the one or more selected image data feature types from the set of non-training SDR image data features of the one or more selected image data feature types extracted from the one or more non-training images;
causing the one or more prediction models to use the set of non-training HDR image data features of the one or more selected image data feature types to generate the one or more optimized values for the one or more image metadata parameters for the one or more non-training images.
2. The method of claim 1 , wherein the one or more prediction models are used to predict the one or more optimized values for the one or more image metadata parameters for one or more non-training images.
3. The method of claim 1 , wherein the one or more selected image data feature types includes one or more of: three-dimensional (3D) histograms comprising 3D cubes with pixel counts, or one-dimensional (1D) luma chroma combined histograms comprising 1D bins with pixel counts.
4. The method of claim 1 , wherein the one or more prediction models include one or more Gaussian process regression (GPR) based prediction models.
5. The method of claim 1 , wherein the one or more image metadata parameters represent one or more display management parameters.
6. The method of claim 1 , wherein the one or more image metadata parameters represent a set of slope, offset and power parameters used to adjust a tone mapping curve for performing display management.
7. The method of claim 1 , wherein at least one of the one or more non-training images is represented in one of: an IPT PQ (ICtCp) color space, an YCbCr color space, an RGB color space, a Rec. 2020 color space, a Rec. 709 color space, an extended dynamic range (EDR) color space, a gamma/HLG/PQ color space, or a standard dynamic range (SDR) color space.
8. The method of claim 1 , wherein at least one of the one or more non-training images is encoded in a sequence of images of the same dynamic range in a video signal of a bit depth of one of 8, 9, 10, 11, 12, 13, 14, or 15+ bits.
9. The method of claim 1 , wherein the plurality of sets of image data features of one or more selected image data feature types is represented by histograms with a plurality of bin partitions to cover an entire range of codeword values in each color channel of a color space in which the plurality of sets of training images is represented.
10. An apparatus comprising a processor and configured to perform the methods recited in claim 1 .
11. A non-transitory computer-readable storage medium having stored thereon computer-executable instruction for executing a method in accordance with the method recited in claim 1 .