Abstract: Systems and methods are disclosed for in-loop, region-based, reshaping for the coding of high-dynamic range video. Using a high bit-depth buffer to store input data and previously decoded reference data, forward and backward, in-loop, reshaping functions allow video coding and decoding to be performed at a target bit depth lower than the input bit depth. Methods for the clustering of the reshaping functions to reduce data overhead are also presented.

Abstract: Noise levels in pre-reshaped codewords of a pre-reshaped bit depth in pre-reshaped images within a time window of a scene are calculated. Per-bin minimal bit depth values are computed for pre-reshaped codeword bins based on the calculated noise levels in the pre-reshaped codewords. Each per-bin minimal bit depth value corresponds to a minimal bit depth value for a respective pre-reshaped codeword bin. A specific codeword mapping function for a specific pre-reshaped image in the pre-reshaped image is generated based on the pre-reshaped bit depth, the per-bin minimal bit depth values, and a target bit depth smaller than the pre-reshaped bit depth. The specific codeword mapping function is applied to specific pre-reshaped codewords of the specific pre-reshaped image to generate specific target codewords of the target bit depth for a specific output image.

Abstract: Methods for screen-adaptive decoding of video with high dynamic range (HDR) are described. The methods combine the traditional compositing and display management steps into one screen-adaptive compositing step. Given decoded standard dynamic range (SDR) input data, metadata related to the prediction of output HDR data in a reference dynamic range, and the dynamic range of a target display, new output luma and chroma prediction functions are generated that map directly the input SDR data to output HDR data in the target dynamic range, thus eliminating the display management step.

Abstract: An SDR CDF is constructed based on an SDR histogram generated from a distribution of SDR codewords in SDR images. An HDR CDF is constructed based on an HDR histogram generated from a distribution of HDR codewords in HDR images that correspond to the SDR images. A histogram transfer function is generated based on the SDR CDF and the HDR CDF. The SDR images are transmitted along with backward reshaping metadata to recipient devices. The backward reshaping metadata is generated at least in part on the histogram transfer function.

Abstract: Compared to traditional gamma-coded video, perceptually quantized video provides greater flexibility for the transmission and display management of high-dynamic range video, but it does not compresses as efficiently using existing standard codecs. Techniques are described to improve the coding efficiency of perceptually coded video by applying a color cross-talk transformation after the RGB/XYZ to LMS transformation. Such a transform increases luma and chroma correlation for color appearance models, but improves perceptual uniformity and overall coding efficiency for wide color gamut, HDR, signals.

Abstract: Techniques use multiple lower bit depth (e.g., 8 bits) codecs to provide higher bit depth (e.g., 12+ bits) high dynamic range images from an upstream device to a downstream device. Multiple layers comprising a base layer and one or more enhancement layers may be used to carry video signals comprising image data compressed by lower bit depth encoders to a downstream device, wherein the base layer cannot be decoded and viewed on its own. Lower bit depth input image data to base layer processing may be generated from higher bit depth high dynamic range input image data via advanced quantization to minimize the volume of image data to be carried by enhancement layer video signals. The image data in the enhancement layer video signals may comprise residual values, quantization parameters, and mapping parameters based in part on a prediction method corresponding to a specific method used in the advanced quantization.

Abstract: A set of optimized operational parameter values is generated for performing decontouring operations on a predicted image. The predicted image is predicted from a first image mapped from a second image that has a higher dynamic range than the first image. Based on the set of optimized operational parameter values, smoothen operations and selection/masking based on a residual mask are performed on the predicted image. The set of optimized operational parameter values is encoded into a part of a multi-layer video signal that includes the first image, and can be used by a recipient decoder to generate a decontoured image based on the predicted image and reconstruct a version of the second image.

Abstract: A content-adaptive quantizer processor receives an input image with an input bit depth. A noise-mask generation process is applied to the input image to generate a noise mask image which characterizes each pixel in the input image in terms of its perceptual relevance in masking quantization noise. A noise mask histogram is generated based on the input image and the noise mask image. A masking-noise level to bit-depth function is applied to the noise mask histogram to generate minimal bit depth values for each bin in the noise mask histogram. A codeword mapping function is generated based on the input bit depth, a target bit depth, and the minimal bit depth values. The codeword mapping function is applied to the input image to generate an output image in the target bit depth.

Abstract: A sequence of visual dynamic range (VDR) images is encoded using a standard dynamic range (SDR) base layer and one or more enhancement layers. A predicted VDR image is generated from an SDR input by using a weighted, multi-band, cross-color channel prediction model. Exponential weights with an adaptable decay parameter for each band are also presented.

Abstract: Statistical values are computed based on received source images. An adaptive reshaping function is selected for one or more source images based on the one or more statistical values. A portion of source video content is adaptively reshaped, based on the selected adaptive reshaping function to generate a portion of reshaped video content. The portion of source video content is represented by the one or more source images. An approximation of an inverse of the selected adaptive reshaping function is generated. The reshaped video content and a set of adaptive reshaping parameters defining the approximation of the inverse of the selected adaptive reshaping function are encoded into a reshaped video signal. The reshaped video signal may be processed by a downstream recipient device to generate a version of reconstructed source images, for example, for rendering with a display device.

Abstract: The present invention relates generally to images. More particularly, an embodiment of the present invention relates to the pixel group segmented quantization and de-quantization of the residual signal in layered coding of high dynamic range images. By assigning the pixels in the residual image to different pixel groups based on the pixel value of the corresponding pixel in the decoded base layer signal, and by applying pixel group quantizing functions to assigned pixels a more efficient coding can be achieved.

Abstract: An encoder receives one or more input pictures of enhanced dynamic range (EDR) to be encoded in a coded bit stream comprising a base layer and one or more enhancement layers. To encode the chroma pixels, the encoder generates a luma mask and a corresponding chroma mask. Based on generated high-clipping and low-clipping thresholds, the encoder determines the appropriate parameters to encode the chroma values in the base and enhancement layers.

Abstract: A scene change is determined using a first and a second video signal, each representing the same scene or content, but at a different color grade (such as dynamic range). A set of prediction coefficients is generated to generate prediction signals approximating the first signal based on the second signal and a prediction model. A set of prediction error signals is generated based on the prediction signals and the first signal. Then, a scene change is detected based on the characteristics of the prediction error signals. Alternatively, a set of entropy values of the difference signals between the first and second video signals are computed, and a scene change is detected based on the characteristics of the entropy values.

Abstract: In a system for coding high dynamic range (HDR) images using lower-dynamic range (LDR) images, a reshaping function allows for a more efficient distribution of the codewords in the lower dynamic range images for improved compression. A trim pass of the LDR images by a colorist may satisfy a director's intent for a given “look,” but may also result in unpleasant clipping artifacts in the reconstructed HDR images. Given an original forward reshaping function which maps HDR luminance values to LDR pixel values, a processor identifies areas of potential clipping and generates modified forward and backward reshaping functions to reduce the visibility of potential artifacts from the trim pass process while preserving the director's intent.

Abstract: A processor for approximating a reshaping function using a multi-segment polynomial receives an input reshaping function. Given a number of target segments (N) and an initial maximum fitting error, in a first pass, it applies a first smoothing filter to the input reshaping function to generate a first smoothed reshaping function. Next, it generates a first multi-segment polynomial approximation of the input reshaping function based on one or more multi-segment polynomial approximation algorithms, the smoothed reshaping function, the number of target segments, and the initial maximum fitting error. The same process may be repeated in two or more similar passes that may include in each pass: reconstructing the reshaping function from the polynomial approximation of the previous pass, smoothing and segmenting the reconstructed reshaping function, and generating an updated multi-segment polynomial approximation according to an updated maximum fitting error.

Abstract: A processor for video coding receives a full-frame rate (FFR) HDR video signal and a corresponding FFR SDR video signal. An encoder generates a scalable bitstream that allows decoders to generate half-frame-rate (HFR) SDR, FFR SDR, HFR HDR, or FFR HDR signals. Given odd and even frames of the input FFR SDR signal, the scalable bitstream combines a base layer of coded even SDR frames with an enhancement layer of coded packed frames, where each packed frame includes a downscaled odd SDR frame, a downscaled even HDR residual frame, and a downscaled odd HDR residual frame. In an alternative implementation, the scalable bitstream combines four signals layers: a base layer of even SDR frames, an enhancement layer of odd SDR frames, a base layer of even HDR residual frames and an enhancement layer of odd HDR residual frames. Corresponding decoder architectures are also presented.

Abstract: A first signal in a first color format is received. The first signal is transformed to a second signal in a second color format to be compressed using an encoder. Before encoding, the chroma components of the second signal are reshaped according to the statistical characteristics of the chroma components of the first signal to generate a third signal. In a decoder, an inverse reshaping functions is applied to the chroma components of the decoded signal to generate an approximation of the second signal.

Abstract: Inter-color image prediction is based on multi-channel multiple regression (MMR) models. Image prediction is applied to the efficient coding of images and video signals of high dynamic range. MMR models may include first order parameters, second order parameters, and cross-pixel parameters. MMR models using extension parameters incorporating neighbor pixel relations are also presented. Using minimum means-square error criteria, closed form solutions for the prediction parameters are presented for a variety of MMR models.

Abstract: Compared to traditional gamma-coded video, perceptually quantized video provides greater flexibility for the transmission and display management of high-dynamic range video, but it does not compresses as efficiently using existing standard codecs. Techniques are described to improve the coding efficiency of perceptually coded video by applying a color cross-talk transformation after the RGB/XYZ to LMS transformation. Such a transform increases luma and chroma correlation for color appearance models, but improves perceptual uniformity and overall coding efficiency for wide color gamut, HDR, signals.

Abstract: A sparse FIR filter can be used to process an image in order to rectify imaging artifacts. In a first example application, the sparse FIR filter is applied as a selective sparse FIR filter that examines a set of selected neighboring pixels of an original pixel in order to identify smooth areas of the image and to selectively apply filtering to only the smooth areas of the image. The parameters of selective filtering are selected based on the characteristics of an inter-layer predictor. In a second example application, the sparse FIR filter is applied as an edge aware selective sparse FIR filter that examines additional neighboring pixels to the set of selected pixels in order to identify edges and carry out selective filtering of smooth areas of the image. Examples for detecting and removing banding artifacts during the coding of high-dynamic range images are provided.