US 12,086,969 B2
Machine learning based dynamic composing in enhanced standard dynamic range video (SDR+)
Harshad Kadu, Santa Clara, CA (US); Neeraj J. Gadgil, San Jose, CA (US); and Guan-Ming Su, Fremont, CA (US)
Assigned to DOLBY LABORATORIES LICENSING CORPORATION, San Francisco, CA (US)
Appl. No. 17/415,650
Filed by DOLBY LABORATORIES LICENSING CORPORATION, San Francisco, CA (US)
PCT Filed Dec. 16, 2019, PCT No. PCT/US2019/066595
§ 371(c)(1), (2) Date Jun. 17, 2021,
PCT Pub. No. WO2020/131731, PCT Pub. Date Jun. 25, 2020.
Claims priority of provisional application 62/781,185, filed on Dec. 18, 2018.
Claims priority of application No. 18213670 (EP), filed on Dec. 18, 2018.
Prior Publication US 2022/0058783 A1, Feb. 24, 2022
Int. Cl. G06T 5/92 (2024.01); G06N 20/00 (2019.01); G06T 7/90 (2017.01)
CPC G06T 5/92 (2024.01) [G06N 20/00 (2019.01); G06T 7/90 (2017.01); G06T 2207/20081 (2013.01); G06T 2207/20208 (2013.01)] 15 Claims
OG exemplary drawing
 
1. A method comprising:
receiving a plurality of training image pairs comprising a plurality of training standard dynamic range (SDR) image and a plurality of corresponding training high dynamic range (HDR) images, wherein each training image pair in the plurality of training image pairs comprises a training SDR image in the plurality of training SDR images and a corresponding training HDR image in the plurality of corresponding training HDR images, wherein the training SDR image and the corresponding training HDR image in each such training image pair depict same visual content but with different luminance dynamic ranges;
extracting a plurality of training image feature vectors from a plurality of training SDR images in the plurality of training image pairs, wherein a training image feature vector in the plurality of training image feature vectors is extracted from a training SDR image in a respective training image pair in the plurality of training image pairs;
using the plurality of training image feature vectors and ground truth derived with the plurality of corresponding training HDR images to train one or more backward reshaping metadata prediction models for predicting operational parameter values of backward reshaping mappings used to backward reshape SDR images into mapped HDR images; and
applying the one or more backward reshaping metadata prediction models to generate a set of operational parameter values specifying an image-specific luma backward reshaping curve used to backward reshape SDR luma codewords of the SDR images into mapped HDR luma codewords of the mapped HDR images.