Dolby Laboratories Licensing Corporation

Neeraj J Gadgil

Guan-Ming Su

Qing Song

Tao Chen

Yoon Yung Lee

Peng Yin

Sheng Qu

Harshad Kadu

Qian Chen

Qiang Zhu

Bihan Wen

Arun Raj

Yee Jin Lee

Training image pairs comprising training SDR image and corresponding training HDR images are received. Each training image pair in the training image pairs comprises a training SDR image and a corresponding training HDR image. The training SDR image and the corresponding training HDR image in the training image pair depict same visual content but with different luminance dynamic ranges. Training image feature vectors are extracted from training SDR images in the training image pairs. The training image feature vectors are used to train backward reshaping metadata prediction models for predicting operational parameter values of backward reshaping mappings used to backward reshape SDR images into mapped HDR images.

Machine learning based dynamic composing in enhanced standard dynamic range video (SDR+)

Methods and systems for reducing banding artifacts when displaying images are described. Identified image bands are filtered using an adaptive sparse finite response filter, where the tap-distance in the sparse filter is adapted according to an estimated width of each image band. Image debanding may be performed across multiple pixel orientations, such as rows, columns, a 45-degree angle, or a &#x2212;45-degree angle. Given a threshold to decide whether sparse filtering needs to be performed or not, an iterative debanding process is also proposed.

Image debanding using adaptive sparse filtering

Methods and systems for image denoising when displaying high-dynamic-range images are described. Given an input image in a first dynamic range, and an input backward reshaping function mapping codewords from the first dynamic range to a second dynamic range, wherein the second dynamic range is equal or higher than the first dynamic range, statistical data based on the input image and the input backward reshaping function are generated to estimate the risk of noise artifacts in a target image in the second dynamic range generated by applying the input backward reshaping function to the input image. Using a measure of the variance in codeword bins in histograms of consecutive input frames (denoted as temporal histogram variance), a modified backward reshaping function is generated, which when applied to the input image to generate the target image eliminates or reduces noise artifacts in the target image.

Image denoising in SDR to HDR image conversion

Methods and systems for reducing banding artifacts when displaying high-dynamic-range images are described. Given an input image in a first dynamic range, and an input backward reshaping function mapping codewords from the first dynamic range to a second dynamic range, wherein the second dynamic range is equal or higher than the first dynamic range, statistical data based on the input image and the input backward reshaping function are generated to estimate the risk of banding artifacts in a target image in the second dynamic range generated by applying the input backward reshaping function to the input image. Separate banding alleviation algorithms are applied in the darks and highlights parts of the first dynamic range to generate a modified backward reshaping function, which when applied to the input image to generate the target image eliminates or reduces banding in the target image.

Reducing banding artifacts in HDR imaging via adaptive SDR-to-HDR reshaping functions

In some embodiments, an encoder device is disclosed to receive an input video stream containing images in a first dynamic range including a first image. The device receives a second image representing the first image. The device obtains statistical data for the first and the second images. The device determines, at a first time delay, a scene cut data from the input video stream and storing the scene cut data in a first sliding window. The device determines, at a second time delay, a first smoothing mapping function based on a second sliding window and the determined scene cut data. The device determines, at a third time delay, a second smoothing mapping function based on a third sliding window and the determined scene cut data. The device generates, at the third time delay, a composer metadata for the first image based on the first and second smoothing mapping functions.

Linear encoder for image/video processing

A standard dynamic range (SDR) image is received. Composer metadata is generated for mapping the SDR image to an enhanced dynamic range (EDR) image. The composer metadata specifies a backward reshaping mapping that is generated from SDR-EDR image pairs in a training database. The SDR-EDR image pairs comprise SDR images that do not include the SDR image and EDR images that corresponds to the SDR images. The SDR image and the composer metadata are encoded in an output SDR video signal. An EDR display operating with a receiver of the output SDR video signal is caused to render an EDR display image. The EDR display image is derived from a composed EDR image composed from the SDR image based on the composer metadata.

Efficient end-to-end single layer inverse display management coding

Methods and systems for reducing banding artifacts when displaying high-dynamic-range images reconstructed from coded reshaped images are described. Given an input image in a high dynamic range (HDR) which is mapped to a second image in a second dynamic range, banding artifacts in a reconstructed HDR image generated using the second image are reduced by a) in darks and mid-tone regions of the input image, adding noise to the input image before being mapped to the second image, and b) in highlights regions of the input image, modifying an input backward reshaping function, wherein the modified backward reshaping function will be used by a decoder to map a decoded version of the second image to the reconstructed HDR image. An example noise generation technique using simulated film-grain noise is provided.

Reducing banding artifacts in backward-compatible HDR imaging

Sequence-level parameters are generated for an image frame sequence including sequence-level indicators for indicating metadata types present for each image frame in the sequence of image frames. Frame-present parameters are generated for a specific image frame in the sequence including frame-present indicators corresponding to the metadata types as indicated in the sequence-level parameters. The frame-present indicators identify first metadata types for which metadata parameter values are to be encoded in a coded bitstream as metadata payloads. The image frame sequence, the sequence-level parameters, the frame-present parameters and the metadata payloads are encoded in the coded bitstream. A recipient device can generate, from the specific image frame based partly on the metadata parameter values determined for the first metadata types, a target display image for a target display.

Backward compatible display management metadata compression

A standard dynamic range (SDR) image and a reference backward reshaping mapping are received. The reference backward reshaping mapping comprises a reference luma backward reshaping mapping. A color preservation mapping function is used with inputs generated from the SDR image and the reference backward reshaping mapping to determine luminance increase for SDR luma histogram bins generated based on luma codewords in the SDR image. A modified backward reshaping mapping is generated and comprises a modified luma backward reshaping mapping generated from the reference backward reshaping function based on the luminance increase for the SDR luma histogram bins. The SDR image and the modified backward reshaping mapping are encoded into an SDR video signal.

Color appearance preservation in video codecs

Given a standard-dynamic range (SDR) video input, techniques for generating and compressing composer metadata describing inverse luma and chroma reshaping functions are described. Given the SDR input, the composer metadata allow a decoder to generate a corresponding output in high-dynamic range. Three techniques are proposed: a static, sequence-based, architecture, a two-stage, scene-based, distributed solution with a centralized post-processing method, and a single-stage distributed solution using overlapped segments. Techniques to reduce the amount of transmitted composer metadata are also described.

Growing community of inventors

Sunnyvale, CA, United States of America

Neeraj J Gadgil