WO2024035939A1 - Procédé et appareil de prédiction entre composantes pour un codage vidéo - Google Patents

Procédé et appareil de prédiction entre composantes pour un codage vidéo Download PDF

Info

Publication number
WO2024035939A1
WO2024035939A1 PCT/US2023/030091 US2023030091W WO2024035939A1 WO 2024035939 A1 WO2024035939 A1 WO 2024035939A1 US 2023030091 W US2023030091 W US 2023030091W WO 2024035939 A1 WO2024035939 A1 WO 2024035939A1
Authority
WO
WIPO (PCT)
Prior art keywords
luma sample
sample value
luma
value
operated
Prior art date
Application number
PCT/US2023/030091
Other languages
English (en)
Inventor
Che-Wei Kuo
Hong-Jheng Jhu
Xiaoyu XIU
Ning Yan
Wei Chen
Bing Yu
Xianglin Wang
Original Assignee
Beijing Dajia Internet Information Technology Co., Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Dajia Internet Information Technology Co., Ltd filed Critical Beijing Dajia Internet Information Technology Co., Ltd
Publication of WO2024035939A1 publication Critical patent/WO2024035939A1/fr

Links

Classifications

    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/169—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding
    • H04N19/186—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the coding unit, i.e. the structural portion or semantic portion of the video signal being the object or the subject of the adaptive coding the unit being a colour or a chrominance component
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/102—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or selection affected or controlled by the adaptive coding
    • H04N19/103—Selection of coding mode or of prediction mode
    • H04N19/11—Selection of coding mode or of prediction mode among a plurality of spatial predictive coding modes
    • H—ELECTRICITY
    • H04—ELECTRIC COMMUNICATION TECHNIQUE
    • H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
    • H04N19/00—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals
    • H04N19/10—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding
    • H04N19/134—Methods or arrangements for coding, decoding, compressing or decompressing digital video signals using adaptive coding characterised by the element, parameter or criterion affecting or controlling the adaptive coding
    • H04N19/136—Incoming video signal characteristics or properties

Definitions

  • aspects of the present disclosure relate generally to image/video coding and compression, and more particularly, to methods and apparatus for cross-component prediction technology.
  • Video coding is performed according to one or more video coding standards.
  • video coding standards include versatile video coding (VVC), high-efficiency video coding (H.265/HEVC), advanced video coding (H.264/AVC), moving picture expert group (MPEG) coding, or the like.
  • Video coding generally utilizes prediction methods (e.g., inter-prediction, intra- prediction, or the like) that take advantage of redundancy present in video images or sequences.
  • An important goal of video coding techniques is to compress video data into a form that uses a lower bit rate, while avoiding or minimizing degradations to video quality.
  • a method for decoding video data comprising: obtaining a video block from a bitstream; obtaining internal luma sample values of the video block, external luma sample values of an external region of the video block and external chroma sample values of the external region; determining, based on the external luma sample values and the external chroma sample values, a set of weighting coefficients corresponding to a filter shape, wherein the filter shape and the set of weighting coefficients are configured for predicting an chroma sample value based on a plurality of corresponding luma sample values, and wherein the set of weighting coefficients includes a weighting coefficient for a non-linear value derived by scaling the product of two of a plurality of corresponding luma sample values or a square of one of the plurality of corresponding luma sample values to a luma sample value range of the video block, predicting, with the filter shape and the set of weighting coefficients, the internal
  • a method for encoding video data comprising: obtaining a video block; obtaining internal luma sample values of the video block, external luma sample values of an external region of the video block and external chroma sample values of the external region; determining, based on the external luma sample values and the external chroma sample values, a set of weighting coefficients corresponding to a filter shape, wherein the filter shape and the set of weighting coefficients are configured for predicting an chroma sample value based on a plurality of corresponding luma sample values, and wherein the set of weighting coefficients includes a weighting coefficient for a non-linear value derived by scaling the product of two of a plurality of corresponding luma sample values or a square of one of the plurality of corresponding luma sample values to a luma sample value range of the video block; predicting, with the filter shape and the set of weighting coefficients, the internal chroma
  • a computer system comprising: one or more processors; and one or more storage devices storing computer-executable instructions that, when executed, cause the one or more processors to perform the operations of the method of the present disclosure.
  • a computer program product storing computer-executable instructions that, when executed, cause one or more processors to perform the operations of the method of the present disclosure.
  • a computer readable medium storing computer-executable instructions that, when executed, cause one or more processors to perform the operations of the method of the present disclosure based on the bitstream.
  • a computer readable medium storing computer-executable instructions that, when executed, cause one or more processors to perform the operations of the method of the present disclosure.
  • a computer readable medium storing a bitstream, wherein the bitstream is to be decoded by performing the operations of the method of the present disclosure.
  • a computer readable medium storing a bitstream, wherein the bitstream is obtained by performing the operations of the method of the present disclosure.
  • Figure 1 illustrates a block diagram of a generic block-based hybrid video encoding system.
  • Figures 2A to 2E illustrate five splitting types, comprising quaternary partitioning, horizontal binary partitioning, vertical binary partitioning, horizontal ternary partitioning, and vertical ternary partitioning.
  • Figure 3 illustrates a general block diagram of a block-based video decoder.
  • Figure 4 illustrates an example of the locations of the left and above samples and the sample of the current block involved in the CCLM mode.
  • Figures 5A to 5C illustrate examples of deriving CCLM parameters.
  • Figure 6 illustrates an example of classifying the neighboring samples into two groups based on the value Threshold.
  • Figure 7 illustrates an example of classifying the neighboring samples into two groups based on a knee point.
  • Figures 8A and 8B illustrate the effect of the scale adjustment parameter “u”.
  • Figure 8C illustrates the collocated reconstructed luma samples.
  • Figure 8D illustrates the neighboring reconstructed samples.
  • Figures 8E to 8H illustrate the steps of decoder-side intra mode derivation.
  • Figure 9 illustrates an example of four reference lines neighboring to a prediction block.
  • Figure 10A illustrates an exemplary pattern for convolutional cross-component model (CCCM).
  • Figure 10B illustrates an exemplary reference area which consists of 6 lines of chroma samples above and left of the PU.
  • Figures 10C and 10D illustrate schematic diagrams for correlation among a chroma sample and one or more luma samples.
  • FIG 11 illustrates an example that 6-tap is used in multiple linear regression (MLR) model according to one or more aspects of the present disclosure.
  • Figure 12 illustrates exemplary different fdter shapes and/or numbers of taps according to one or more aspects of the present disclosure.
  • Figure 13 illustrates an example in which FLM can only use top or left luma and/or chroma samples (extended) for parameter derivation.
  • Figure 14 illustrates an example in which FLM can use different lines for parameter derivation.
  • Figures 15A to 15D illustrate some examples for l-tap/2-tap pre-operations.
  • Figure 16 illustrates examples of different shape/number of filter taps.
  • Figure 17 illustrates examples of different shape/number of filter taps.
  • Figures 18 A and 18B illustrate examples of different shape/number of filter taps.
  • Figures 19A to 19G illustrate examples of different set of filter taps.
  • Figure 20A and Figure 20B illustrate examples of 2-fold training for implicitly filter shape derivation.
  • Figure 21 illustrates a workflow of a method for decoding video data according to one or more aspects of the present disclosure.
  • Figure 22 illustrates a workflow of a method for encoding video data according to one or more aspects of the present disclosure.
  • Figure 23 illustrates an exemplary computing system according to one or more aspects of the present disclosure.
  • the first version of the VVC standard was finalized in July, 2020, which offers approximately 50% bit-rate saving or equivalent perceptual quality compared to the prior generation video coding standard HEVC.
  • the VVC standard provides significant coding improvements than its predecessor, there is evidence that superior coding efficiency can be achieved with additional coding tools.
  • Joint Video Exploration Team JVET
  • ISO/IEC MPEG started the exploration of advanced technologies that can enable substantial enhancement of coding efficiency over VVC.
  • ECM Enhanced Compression Model
  • VTM VVC Test Model
  • CTCs JVET common test conditions
  • FIG. 1 illustrates a block diagram of a generic block-based hybrid video encoding system.
  • the input video signal is processed block by block (called coding units (CUs)).
  • CUs coding units
  • a CU can be up to 128x128 pixels.
  • one coding tree unit (CTU) is split into CUs to adapt to varying local characteristics based on quad/binary/ternary-tree.
  • CTU coding tree unit
  • the multi-type tree structure one CTU is firstly partitioned by a quad-tree structure.
  • each quad-tree leaf node can be further partitioned by a binary and ternary tree structure.
  • FIGs 2A, 2B, 2C, 2D, and 2E there are five splitting types, quaternary partitioning, vertical binary partitioning, horizontal binary partitioning, vertical extended quaternary partitioning, and horizontal extended quaternary partitioning.
  • spatial prediction and/or temporal prediction may be performed.
  • Spatial prediction (or “intra prediction”) uses pixels from the samples of already coded neighboring blocks (which are called reference samples) in the same video picture/slice to predict the current video block. Spatial prediction reduces spatial redundancy inherent in the video signal.
  • Temporal prediction (also referred to as “inter prediction” or “motion compensated prediction”) uses reconstructed pixels from the already coded video pictures to predict the current video block. Temporal prediction reduces temporal redundancy inherent in the video signal.
  • Temporal prediction signal for a given CU is usually signaled by one or more motion vectors (MVs) which indicate the amount and the direction of motion between the current CU and its temporal reference.
  • MVs motion vectors
  • one reference picture index is additionally sent, which is used to identify from which reference picture in the reference picture store the temporal prediction signal comes.
  • the mode decision block in the encoder chooses the best prediction mode, for example based on the rate-distortion optimization method.
  • the prediction block is then subtracted from the current video block; and the prediction residual is de-correlated using transform and quantized.
  • the quantized residual coefficients are inverse quantized and inverse transformed to form the reconstructed residual, which is then added back to the prediction block to form the reconstructed signal of the CU.
  • in-loop filtering such as deblocking filter, sample adaptive offset (SAO) and adaptive in-loop filter (ALF) may be applied on the reconstructed CU before it is put in the reference picture store and used to code future video blocks.
  • coding mode inter or intra
  • prediction mode information motion information
  • quantized residual coefficients are all sent to the entropy coding unit to be further compressed and packed to form the bit-stream.
  • block or “video block” as used herein may be a portion, in particular a rectangular (square or non- square) portion, of a frame or a picture.
  • the block or video block may be or correspond to a Coding Tree Unit (CTU), a CU, a Prediction Unit (PU) or a Transform Unit (TU) and/or may be or correspond to a corresponding block, e.g., a Coding Tree Block (CTB), a Coding Block (CB), a Prediction Block (PB) or a Transform Block (TB) and/or to a sub-block.
  • CTU Coding Tree Block
  • PU Prediction Unit
  • TU Transform Unit
  • a corresponding block e.g., a Coding Tree Block (CTB), a Coding Block (CB), a Prediction Block (PB) or a Transform Block (TB) and/or to a sub-block.
  • CTB Coding Tree Block
  • PB Prediction Block
  • TB Transform Block
  • Figure 3 illustrates a general block diagram of a block-based video decoder.
  • the video bit-stream is first entropy decoded at entropy decoding unit.
  • the coding mode and prediction information are sent to either the spatial prediction unit (if intra coded) or the temporal prediction unit (if inter coded) to form the prediction block.
  • the residual transform coefficients are sent to inverse quantization unit and inverse transform unit to reconstruct the residual block.
  • the prediction block and the residual block are then added together.
  • the reconstructed block may further go through in-loop filtering before it is stored in reference picture store.
  • the reconstructed video in reference picture store is then sent out to drive a display device, as well as used to predict future video blocks.
  • the main focus of this disclosure is to further enhance the coding efficiency of the coding tool of cross-component prediction, cross-component linear model (CCLM), that is applied in the ECM.
  • CCLM cross-component linear model
  • some related coding tools in the ECM are briefly reviewed. After that, some deficiencies in the existing design of CCLM are discussed. Finally, the solutions are provided to improve the existing CCLM prediction design.
  • the above a and p are linear model parameters which are derived from at most four neighboring chroma samples and their corresponding down-sampled luma samples, which may be referred to as neighboring luma-chroma sample pairs.
  • positions of four neighboring chroma samples are selected as follows:
  • the four neighboring luma samples corresponding to the selected locations are obtained by a down-sampling operation and the obtained four neighboring luma samples are compared four times to find two larger values: x 0 A and x 1 A , and two smaller values: x 0 B and x 1 B .
  • Chroma sample values corresponding to the two larger values and the two smaller values are denoted as Y 0 A, Y 1 A, Y 0 B and Y 1 B respectively.
  • X a , X b , Y a and Y b are derived as:
  • Figure 4 illustrates an example of the locations of the left and above samples and the sample of the current block involved in the CCLM mode, including locations of left and above samples of an N X N chroma block in the CU and locations of left and above samples of an 2N X 2N luma block in the CU.
  • the division operation to calculate parameter a is implemented with a look-up table.
  • the diff value difference between maximum and minimum values
  • the parameter a are expressed by an exponential notation. For example, diff is approximated with a 4-bit significant part and an exponent. Consequently, the table for 1/diff is reduced into 16 elements for 16 values of the significand as follows:
  • LM_A 2 LM modes
  • LM_L 2 LM modes
  • LM_T mode only the above template is used to calculate the linear model coefficients. To get more samples, the above template is extended to (W+H) samples.
  • LM_L mode only left template is used to calculate the linear model coefficients. To get more samples, the left template is extended to (H+W) samples.
  • two types of down-sampling filter are applied to luma samples to achieve 2 to 1 down-sampling ratio in both horizontal and vertical directions.
  • the selection of down-sampling filter is specified by a SPS level flag.
  • the two down-sampling filters are as follows, which are corresponding to “type-0” and “type-2” content, respectively.
  • This parameter computation is performed as part of the decoding process, and is not just as an encoder search operation. As a result, no syntax is used to convey the ⁇ and ⁇ values to the decoder.
  • Chroma mode signalling and derivation process are shown in Table 1.
  • Chroma mode coding directly depends on the intra prediction mode of the corresponding luma block. Since separate block partitioning structure for luma and chroma components is enabled in I slices, one chroma block may correspond to multiple luma blocks. Therefore, for Chroma DM mode, the intra prediction mode of the corresponding luma block covering the center position of the current chroma block is directly inherited.
  • the first bin indicates whether it is regular (0) or LM modes (1). If it is LM mode, then the next bin indicates whether it is LM CHROMA (0) or not. If it is not LM_CHROMA, next 1 bin indicates whether it is LM_L (0) or LM_A (1). For this case, when sps cclm enabled flag is 0, the first bin of the binarization table for the corresponding intra_chroma_pred_mode can be discarded prior to the entropy coding. Or, in other words, the first bin is inferred to be 0 and hence not coded. This single binarization table is used for both sps cclm enabled flag equal to 0 and 1 cases. The first two bins in Table 2 are context coded with its own context model, and the rest bins are bypass coded.
  • the chroma CUs in 32x32 / 32x16 chroma coding tree node are allowed to use CCLM in the following way:
  • all chroma CUs in the 32x16 chroma node can use CCLM.
  • CCLM is not allowed for chroma CU.
  • FIG. 5A illustrates an example that MDLM works when the block content cannot be predicted from the L-shape reconstructed region.
  • Figure 5B illustrates MDLM_L which only uses left reconstructed samples to derive CCLM parameters.
  • Figure 5C illustrates
  • MDLM T which only uses top reconstructed samples to derive CCLM parameters.
  • the integerization design utilizes the linear relationship to modelize the correlation of luma signal and chroma signal.
  • the chroma values are predicted from reconstructed luma values of collocated block.
  • Luma and chroma components have different sampling ratios in YUV420 sampling.
  • the sampling ratio of chroma components is half of that of luma component and has 0.5 pixel phase difference in vertical direction.
  • Reconstructed luma needs down-sampling in vertical direction and subsample in horizontal direction to match size of chroma signal.
  • the down-sampling may be implemented by:
  • Equation (1) Float point operation is necessary in equation (8) to calculate linear model parameters a to keep high data accuracy.
  • float point multiplication is involved in equation (1) when a is represented by float point value.
  • ⁇ is the rounding value of float point fl and ⁇ 'can be calculated as follows.
  • equation ( 12) can be re-written as following. Where is represented as lookup table with length of 2 nA2 to avoid the division.
  • table size can be further reduced to 32 by up-scaling A 2 when bdepth(A 2 ) ⁇ 6 (e.g. A 2 ⁇ 32).
  • n A is set as 15, to avoid product overflow and keep 16 bits multiplication.
  • parameter ⁇ ' is calculated as follows:
  • an intra prediction mode called LM is applied to predict chroma PU based on a linear model using the reconstruction of the collocated luma PU.
  • the parameters of the linear model consist of slope (a»k) and y-intercept (b), which are derived from the neighboring luma and chroma pixels using the least mean square solution.
  • nS specifies the block size of the current chroma PU
  • P Y ’ is the reconstructed pixel from the colocated luma PU.
  • P Y ’ is the reconstructed neighboring pixel of the co-located luma PU.
  • Equation (19-6) als is a 16-bit signed integer and ImDiv is a 16-bit unsigned integer. Therefore, 16-bit multiplier and 16-bit storage are needed. It is proposed to reduce the bit depth of multipliers to the internal bit depth, as well as the size of the look-up table, as detailed below.
  • Table 4 shows the example of internal bit depth 10.
  • Multi-model LM (MMLM) prediction mode, for which the chroma samples are predicted based on the reconstructed luma samples of the same CU by using two linear models as follows: fpredcfi, it rec/ Ci, j) Threshold
  • predcCi it rec/ Ci, j) > Threshold
  • predcG’ , j) represents the predicted chroma samples in a CU and rec/ (i , j) represents the down-sampled reconstructed luma samples of the same CU.
  • Threshold is calculated as the average value of the neighboring reconstructed luma samples.
  • Figure 6 illustrates an example of classifying the neighboring samples into two groups based on the value Threshold.
  • parameter a and [3,, with i equal to 1 and 2 respectively, are derived from the straight-line relationship between luma values and chroma values from two samples, which are minimum luma sample A (X A , Y A ) and maximum luma sample B (X B , Y B ) inside the group.
  • X A , Y A are the x-coordinate (i.e., luma value) and y-coordinate (i.e., chroma value) value for sample A
  • X B , Y B are the x-coordinate and y-coordinate value for sample B.
  • the linear model parameters a and fl are obtained according to the following equations. ye - VA a — -
  • the two templates also can be used alternatively in the other two MMLM modes, called MMLM A, and MMLM L modes.
  • MMLM A mode only pixel samples in the above template are used to calculate the linear model coefficients. To get more samples, the above template is extended to the size of (W+W).
  • MMLM_L mode only pixel samples in the left template are used to calculate the linear model coefficients. To get more samples, the left template is extended to the size of (H+H).
  • Chroma mode coding For chroma intra mode coding, a total of 11 intra modes are allowed for chroma intra mode coding. Those modes include five traditional intra modes and six cross-component linear model modes (CCLM, LM_A, LM_L, MMLM, MMLM A and MMLM L). Chroma mode signaling and derivation process are shown in Table 6. Chroma mode coding directly depends on the intra prediction mode of the corresponding luma block. Since separate block partitioning structure for luma and chroma components is enabled in I slices, one chroma block may correspond to multiple luma blocks. Therefore, for Chroma DM mode, the intra prediction mode of the corresponding luma block covering the center position of the current chroma block is directly inherited.
  • MMLM and LM modes may also be used together in an adaptive manner.
  • two linear models are as follows: where pred c (i, j) represents the predicted chroma samples in a CU and rec L ’ (i, j) represents the down-sampled reconstmcted luma samples of the same CU. Threshold can be simply determined based on the luma and chroma average values together with their minimum and maximum values.
  • Figure 7 shows an example of classifying the neighboring samples into two groups based on the knee point, T, indicated by an arrow.
  • Linear model parameter ⁇ 1 and ⁇ 1 are derived from the straight-line relationship between luma values and chroma values from two samples, which are minimum luma sample A (X A , Y A ) and the Threshold (X T , Y T ).
  • Linear model parameter ⁇ 2 and ⁇ 2 are derived from the straight-line relationship between luma values and chroma values from two samples, which are maximum luma sample B (X B , Y B ) and the Threshold (X T , Y T ).
  • X A , Y A are the x-coordinate (i.e., luma value) and y-coordinate (i.e., chroma value) value for sample A
  • X B , Y B are the x- coordinate and y-coordinate value for sample B.
  • the linear model parameters ⁇ i and ⁇ i for each group, with i equal to 1 and 2 respectively, are obtained according to the following equations.
  • the two templates also can be used alternatively in the other two MMLM modes, called MMLM A, and MMLM_L modes respectively.
  • MMLM A mode only pixel samples in the above template are used to calculate the linear model coefficients. To get more samples, the above template is extended to the size of (W+W).
  • MMLM_L mode only pixel samples in the left template are used to calculate the linear model coefficients. To get more samples, the left template is extended to the size of (H+H).
  • MMLM multi-model LM modes
  • MMLM multi-model LM modes
  • the condition check is as follows: where BlkSizeThres LM represents the smallest block size of LM modes and BlkSizeThres MM represents the smallest block size of MMLM modes.
  • the symbol d represents a pre-determined threshold value. In one example, d may take a value of 0. In another example, d may take a value of 8.
  • Chroma mode signaling and derivation process are shown in Table 1. It is worth noting that for a given CU, if it is coded under linear model mode, whether it is a conventional single model LM mode or a MMLM mode is determined based on the condition check above. Unlike the case shown in Table 6, there are no separate MMLM modes to be signaled. Chroma mode coding directly depends on the intra prediction mode of the corresponding luma block.
  • one chroma block may correspond to multiple luma blocks. Therefore, for Chroma DM mode, the intra prediction mode of the corresponding luma block covering the center position of the current chroma block is directly inherited.
  • CCLM uses a model with 2 parameters to map luma values to chroma values.
  • mapping function is tilted or rotated around the point with luminance value y r . It is proposed to use the average of the reference luma samples used in the model creation as y r in order to provide a meaningful modification to the model.
  • Figures 8A to 8B illustrate the effect of the scale adjustment parameter “u”, wherein Figure 8A illustrates the model created without the scale adjustment parameter “u”, and Figure 8B illustrates the model created with the scale adjustment parameter “u”.
  • the scale adjustment parameter is provided as an integer between -4 and 4, inclusive, and signaled in the bitstream.
  • the unit of the scale adjustment parameter is l/8th of a chroma sample value per one luma sample value (for 10-bit content).
  • adjustment is available for the CCLM models that are using reference samples both above and left of the block (“LM CHROMA IDX” and “MMLM_CHROMA_IDX”), but not for the “single side” modes. This selection is based on coding efficiency vs. complexity trade-off considerations.
  • the encoder may perform an SATD based search for the best value of the scale update for Cr and a similar SATD based search for Cb. If either one results as a non-zero scale adjustment parameter, the combined scale adjustment pair (SATD based update for Cr, SATD based update for Cb) is included in the list of RD checks for the TU.
  • JVET-Y0092/Z0051 proposed fusion of chroma intra modes.
  • the intra prediction modes enabled for the chroma components in ECM-4.0 are six cross-component linear model (LM) modes including CCLM LT, CCLM L, CCLM T, MMLM_LT, MMLM_L and MMLM T modes, the direct mode (DM), and four default chroma intra prediction modes.
  • the four default modes are given by the list ⁇ 0, 50, 18, 1 ⁇ and if the DM mode already belongs to that list, the mode in the list will be replaced with mode 66.
  • a decoder-side intra mode derivation (DIMD) method for luma intra prediction is included in ECM-4.0.
  • a horizontal gradient and a vertical gradient are calculated for each reconstructed luma sample of the L-shaped template of the second neighboring row and column of the current block to build a Histogram of Gradients (HoG).
  • HoG Histogram of Gradients
  • the two intra prediction modes with the largest and the second largest histogram amplitude values are blended with the Planar mode to generate the final predictor of the current luma block.
  • DIMD chroma decoder-side derived chroma intra prediction mode
  • a fusion of a non-LM mode and the MMLM LT mode DIMD chroma
  • a DIMD chroma mode uses the DIMD derivation method to derive the chroma intra prediction mode of the current block based on the collocated reconstructed luma samples. Specifically, a horizontal gradient and a vertical gradient are calculated for each collocated reconstructed luma sample of the current chroma block to build a HoG, as shown in Figure 8C. Then the intra prediction mode with the largest histogram amplitude values is used for performing chroma intra prediction of the current chroma block.
  • the intra prediction mode derived from the DIMD chroma mode is the same as the intra prediction mode derived from the DM mode, the intra prediction mode with the second largest histogram amplitude value is used as the DIMD chroma mode.
  • a CU level flag is signaled to indicate whether the proposed DIMD chroma mode is applied as shown in Table 7.
  • the two weights, w0 and wl are determined by the intra prediction mode of adjacent chroma blocks and shift is set equal to 2.
  • the DIMD chroma mode and the fusion of chroma intra prediction modes are combined. Specifically, the DIMD chroma mode described in the first embodiment is applied, and for I slices, the DM mode, the four default modes and the DIMD chroma mode can be fused with the MMLM LT mode using the weights described in the second embodiment, while for non-I slices, only the DIMD chroma mode can be fused with the MMLM_LT mode using equal weights.
  • the DIMD chroma mode with reduced processing and the fusion of chroma intra prediction modes are combined. Specifically, the DIMD chroma mode with reduced processing derives the intra mode based on the neighboring reconstructed Y, Cb and Cr samples in the second neighboring row and column as shown in Figure 8D. Other parts are the same as the third embodiment.
  • two intra modes are derived from the reconstructed neighbor samples, and those two predictors are combined with the planar mode predictor with the weights derived from the gradients as described in J VET-00449.
  • the division operations in weight derivation is performed utilizing the same lookup table (LUT) based integerization scheme used by the CCLM. For example, the division operation in the orientation calculation
  • DivSigTable[16] ⁇ 0, 7, 6, 5 ,5, 4, 4, 3, 3, 2, 2, 1, 1, 1, 1, 1, 0 ⁇ .
  • Derived intra modes are included into the primary list of intra most probable modes (MPM), so the DIMD process is performed before the MPM list is constructed.
  • the primary derived intra mode of a DIMD block is stored with a block and is used for MPM list construction of the neighboring blocks.
  • Figures 8E to 8H illustrate the steps of decoder-side intra mode derivation, wherein intra prediction direction is estimated without intra mode signaling.
  • the first step as shown in Figure 8E includes estimating gradient per sample (for light-grey samples as illustrated in Figure 8E).
  • the second step as shown in Figure 8F includes mapping gradient values to closest prediction direction within [2,66]
  • the third step as shown in Figure 8G includes selecting 2 prediction directions, wherein for each prediction direction, all absolute gradients Gx and Gy of neighboring pixels with that direction are summed up, and top 2 directions are selected.
  • the fourth step as shown in Figure 8H includes enabling weighted intra prediction with the selected directions.
  • Multiple reference line (MRL) intra prediction uses more reference lines for intra prediction.
  • MRL multiple reference line
  • FIG 9 an example of 4 reference lines is depicted, where the samples of segments A and F are not fetched from reconstructed neighboring samples but padded with the closest samples from Segment B and E, respectively.
  • HEVC intra-picture prediction uses the nearest reference line (i.e., reference line 0).
  • reference line 0 the nearest reference line
  • 2 additional lines reference line 1 and reference line 3 are used.
  • the index of selected reference line (mrl idx) is signaled and used to generate intra predictor.
  • reference line idx which is greater than 0, only include additional reference line modes in MPM list and only signal mpm index without remaining mode.
  • the reference line index is signaled before intra prediction modes, and Planar mode is excluded from intra prediction modes in case a nonzero reference line index is signaled.
  • MRL is disabled for the first line of blocks inside a CTU to prevent using extended reference samples outside the current CTU line. Also, PDPC is disabled when additional line is used.
  • MRL mode the derivation of DC value in DC intra prediction mode for non-zero reference line indices is aligned with that of reference line index 0.
  • MRL requires the storage of 3 neighboring luma reference lines with a CTU to generate predictions.
  • the CrossComponent Linear Model (CCLM) tool also requires 3 neighboring luma reference lines for its down-sampling filters. The definition of MRL to use the same 3 lines is aligned as CCLM to reduce the storage requirements for decoders.
  • CCCM convolutional cross-component model
  • CCCM convolutional cross-component model
  • Multi-model CCCM mode can be selected for PUs which have at least 128 reference samples available.
  • the proposed convolutional 7-tap filter consists of a 5-tap plus sign shape spatial component, a nonlinear term and a bias term.
  • the input to the spatial 5-tap component of the filter consists of a center (C) luma sample which is collocated with the chroma sample to be predicted and its above/north (N), below/south (S), left/west (W) and right/east (E) neighbors as illustrated in Figure 10 A.
  • the nonlinear term P is represented as power of two of the center luma sample C and scaled to the sample value range of the content:
  • the bias term B represents a scalar offset between the input and output (similarly to the offset term in CCLM) and is set to middle chroma value (512 for 10-bit content).
  • Output of the filter is calculated as a convolution between the filter coefficients c; and the input values and clipped to the range of valid chroma samples:
  • predChromaVal c 0 C + C 1 N + c 2 S + c 3 E + c 4 W + c 5 P + c 6 B
  • the filter coefficients Ci are calculated by minimising MSE between predicted and reconstructed chroma samples in the reference area.
  • Figure 10B illustrates the reference area which consists of 6 lines of chroma samples above and left of the PU. Reference area extends one PU width to the right and one PU height below the PU boundaries. Area is adjusted to include only available samples. The extensions to the area shown in blue are needed to support the “side samples” of the plus shaped spatial filter and are padded when in unavailable areas.
  • the MSE minimization is performed by calculating autocorrelation matrix for the luma input and a cross-correlation vector between the luma input and chroma output.
  • Autocorrelation matrix is LDL decomposed and the final filter coefficients are calculated using back-substitution.
  • the process follows roughly the calculation of the ALF filter coefficients in ECM, however LDL decomposition was chosen instead of Cholesky decomposition to avoid using square root operations.
  • the proposed approach uses only integer arithmetic.
  • CCCM Usage of the mode is signalled with a CABAC coded PU level flag.
  • CABAC context was included to support this.
  • CCCM is considered a sub-mode of CCLM. That is, the CCCM flag is only signalled if intra prediction mode is LM CHROMA IDX (to enable single mode CCCM) or MMLM CHROMA IDX (to enable multi-model CCCM).
  • the encoder performs two new RD checks in the chroma prediction mode loop, one for checking single model CCCM mode and one for checking multi-model CCCM mode.
  • the neighboring reconstructed lumachroma sample pairs are classified into one or more sample groups based on the value Threshold, which only considers the luma DC values. That is, a luma-chroma sample pair is classified by only considering the intensity of the luma sample.
  • Threshold which only considers the intensity of the luma sample.
  • luma component usually preserves abundant textures, and the current luma sample may be highly correlated with neighboring luma samples, such inter-sample correlation (AC correlation) may benefit the classification of luma-chroma sample pairs and can bring additional coding efficiency.
  • the CCLM assumes a given chroma sample only correlates to a corresponding luma sample (L0.5, which can be taken as the fractional luma sample position), and a simple linear regression (SLR) with ordinary least squares (OLS) estimation is used to predict the given chroma sample.
  • SLR simple linear regression
  • OLS ordinary least squares
  • one chroma sample may simultaneously correlate to multiple luma samples (AC or DC correlation), so a multiple linear regression (MLR) model may further improve the prediction accuracy.
  • the CCCM mode can enhance the intra prediction efficiency, there is room to further improve its performance. Meanwhile, some parts of the existing CCCM mode also need to be simplified for efficient codec hardware implementations or improved for better coding efficiency. Furthermore, the tradeoff between its implementation complexity and its coding efficiency benefit needs to be further improved.
  • classifiers considering luma edge or AC information is introduced, in contrast to the above implementations wherein only luma DC values are considered.
  • the present disclosure provides exemplary classifiers.
  • the process of generating linear prediction models for different sample groups may be similar as CCLM or MMLM (e.g., via a least square method, or a simplified min-max method, etc.), but with different metrices for classification.
  • Different classifiers may be used to classify the neighboring luma samples (e.g., of the neighboring luma-chroma sample pairs) and/or the luma samples corresponding to chroma samples to be predicted.
  • the luma samples corresponding to the chroma samples may be obtained by a down-sampling operation to match the locations of the corresponding chroma samples for 4:2:0 video sequences.
  • a luma sample corresponding to a chroma sample may be obtained by performing a down-sampling operation on more than one (e.g., 4) reconstructed luma samples corresponding to the chroma sample (e.g., located around the chroma sample).
  • the luma samples may obtained directly from the reconstructed luma samples in a case of 4:4:4 video sequences, for example.
  • the luma samples may be obtained from respective ones of the reconstructed luma samples that are at respective collocated positions for the corresponding chroma samples.
  • a luma sample to be classified may be obtained from one of four reconstructed luma samples corresponding to the chroma sample that is at a left-top position of the four reconstructed luma samples, which may be considered as a collocated position for the chroma sample.
  • a first classifier may classify luma samples according to their edge strengths For example, one direction (e.g., 0-degree, 45-degree, or 90-degree, etc.) may be selected to calculate the edge strength.
  • a direction may be formed by a current sample and a neighboring sample along the direction (e.g., a neighboring sample located at the right-top of the current sample for 45-degree).
  • An edge strength may be calculated by subtracting the neighbor sample from the current sample.
  • the edge strength may be quantized into one of M segments by M-l thresholds, and the first classifier may use M classes to classify the current sample.
  • N directions may be formed by a current sample and N neighboring samples along the N directions.
  • N edge strengths may be calculated by subtracting N neighboring samples from the current sample, respectively. Similarly, if each of the N edge strengths may be quantized into one of M segments by M-l thresholds, then the first classifier may use MN classes to classify the current sample.
  • a second classifier may be used to classify according to a local pattern. For example, a current luma sample Y0 may be compared with its neighboring N luma samples Yi. A score may be added by one if the value of Y0 is greater than that of Yi, otherwise, the score may be reduced by one. The sore may be quantized to form K classes. The second classifier may classify a current sample into one of the K classes. For example, the neighboring luma samples may be obtained from four neighbors that are located above, left, right and below the current luma samples, i.e., without diagonal neighbors.
  • first classifier may be combined with the existing MMLM threshold-based classifier.
  • instance A of the first classifier may be combined with another instance B of the first classifier, where the instance A and B employ different directions (e.g., employing vertical and horizontal directions, respectively).
  • the proposed cross-component method described in the disclosure can also be applied to other prediction coding tools with similar design spirits.
  • the proposed method can also be applied by dividing luma-chroma sample pairs into multiple sample groups.
  • Y/Cb/Cr also can be denoted as Y/U/V in video coding area. If video data is of RGB format, the proposed method can also be applied by simply mapping YUV notation to GBR, for example.
  • a filter-based linear model which utilizes the MLR model is introduced as follows, to take into account the possibilities that one chroma sample may simultaneously correlate to multiple luma samples.
  • the reconstructed collocated and neighboring luma samples can be used to predict the chroma sample, to capture the inter-sample correlation among the collocated luma sample, neighboring luma samples, and the chroma sample.
  • the reconstructed luma samples are linear weighted and combined with one “offset” to generate the predicted chroma sample (C: predicted chroma sample, Li: i -th reconstructed collocated or neighboring luma samples, a filter coefficients, fi : offset, N: filter taps), as shown in the following equation (32-1).
  • the linear weighted plus offset value directly forms the predicted chroma sample (can be low pass, high pass adaptively according to video content), and it is then added by the residual to form the reconstructed chroma sample.
  • the offset term can also be implemented as middle chroma value B (512 for 10-bit content) multiplied by another coefficient, as shown in the following equation (32-2).
  • the top and left reconstructed luma and chroma samples can be used to derive or train the FLM parameters ( a i, , ⁇ ).
  • a . and ⁇ can be derived via OLS.
  • the top and left training samples are collected, and one pseudo inverse matrix is calculated at both encoder and decoder sides to derive the parameters, which are then used to predict the chroma samples in the given CU.
  • N denotes the number of filter taps applied on luma samples
  • M denotes the total top and left reconstructed luma-chroma sample pairs used for training parameters
  • c i denotes the chroma sample with the i-th sample pair
  • the following equations show the derivation of the pseudo inverse matrix A + , and also the parameters.
  • Figure 11 shows an example that N is 6 (6-tap), M is 8, top 2 rows and left 3 columns luma samples and top 1 row and left 1 column chroma samples are used to derive or train the parameters [00159] Please note that one can predict the chroma sample by only a i without the offset ⁇ , which may be a subset of the proposed method.
  • FIG. 11 To further improve the coding performance, additional designs may be used in the FLM prediction. As shown in Figure 11 and discussed above, a 6-tap luma filter is used for the FLM prediction. However, though a multiple tap filter can fit well on training data (e g., top and left neighboring reconstructed luma and chroma samples), in some cases that training data do not capture full characteristics of testing data, it may result in overfitting and may not predict well on testing data (i.e., the to-be-predicted chroma block samples). Also, different filter shapes may adapt well to different video block content, leading to more accurate prediction.
  • training data e g., top and left neighboring reconstructed luma and chroma samples
  • different filter shapes may adapt well to different video block content, leading to more accurate prediction.
  • the filter shape and number of filter taps can be predefined or signaled or switched in Sequence Parameter Set (SPS), Adaptation Parameter Set (APS), Picture Parameter Set (PPS), Picture Header (PH), Slice Header (SH), Region, CTU, CU, Subblock, or Sample level.
  • SPS Sequence Parameter Set
  • APS Adaptation Parameter Set
  • PPS Picture Parameter Set
  • PH Picture Header
  • SH Slice Header
  • Region CTU
  • CU Subblock
  • Sample level Sample level
  • a set of filter shape candidates can be predefined, and a selection on the set of filter shape candidates may be signaled or switched in SPS, APS, PPS, PH, SH, Region, CTU, CU, Subblock, or Sample level.
  • Different components e.g., U and V
  • a set of filter shape candidates may be predefined, and a filter shape (1, 2) may denote a 2-tap luma filter, a filter shape (1, 2, 4) may denote a 3-tap luma filter and the like, as shown in Figure 11.
  • the filter shape selection of U and V components can be switched in PH or in CU or CTU level.
  • N-tap can represent N-tap with or without the offset 0 as described herein.
  • Table 8 One example is given as below in Table 8.
  • the FLM or the CCCM filter shape may include a nonlinear term.
  • the values used to derive the nonlinear term P can be a combination of current and neighboring luma samples, but not limited to C*C.
  • P can be derived as following:
  • Q and R can be linear combination of current and neighboring luma samples either in a down-sampled domain (for example, the Q and R being pre-operated luma samples obtained by weighted-average operation) or without any down-sampling process.
  • Different values (Q and R) used to derive the nonlinear term are considered as different fdter shapes and can be predefined or signaled/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels.
  • a set of filter shape candidates can be predefined or signaled/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels.
  • Different chroma types and/or color formats can have different predefined filter shapes and/or taps.
  • a predefined filter shape (1, 2, 4, 5) may be used for 4:2:0 type-0
  • a predefined filter shape (0, 1, 2, 4, 7) may be used for 4:2:0 type-2
  • a predefined filter shape (1, 4) may be used for 4:2:2
  • a predefined filter shape (0, 1, 2, 3, 4, 5) may be used for 4:4:4, as shown in Figure 12.
  • unavailable luma and chroma samples for deriving the MLR model can be padded from available reconstructed samples. For example, if using a 6-tap (0, 1, 2, 3, 4, 5) filter as in Figure 12, for a CU located at the left picture boundary, the left columns including samples (0, 3) are not available (out of picture boundary), so samples (0, 3) are repetitive padding from samples (1, 4) to apply the 6-tap filter. Note that the padding process may be applied in both training data (top and left neighboring reconstructed luma and chroma samples) and testing data (the luma and chroma samples in the CU(s)).
  • One or more shape/number of filter taps may be used for FLM prediction, examples being shown in Figure 16, Figure 17, and Figures 18A to 18B.
  • One or more sets of filter taps may be used for FLM prediction, examples being shown in Figures 19 A to 19G.
  • the filter shape candidates can be implicitly derived without explicitly signaling bits.
  • the filter shape candidates can be the filter shape candidates for FLM or GLM (as discussed below).
  • the filter shape candidates can be the cross shape filter for CCCM, any of the filters shown in Figure 16, Figure 17, Figure 18A, Figure 18B, and Figures 19Ato 19G, or other filters mentioned in this disclosure.
  • the well- known “N-fold cross-validation” technique in machine learning area can be used for training filter coefficients. The technique divides the available training data into N sets, and use partial sets for training, others for validation.
  • Step 1 determining M filter shape candidates for predicting the chroma sample values of the current CU
  • Step 2 dividing the available L-shaped template area external to the CU into N regions, denoted as R 0 , R 1 ... RN-I , i.e., dividing the training data into N sets for N-fold training, wherein the luma sample values and the chroma sample values of the available template area are known values,
  • Step 3 applying each of the M filter shape candidates respectively to a part of the available template area, i . e. , one or more regions among all the N regions RQ, R .. . RN-I,
  • Step 4 deriving M filter coefficient sets corresponding to the M filter shape candidates, denoted as F o , F b .. . F M.b
  • Step 5 applying the derived F o , F b ... F M -I filter coefficient sets to another part of the available template area to predict the chroma sample values based on corresponding luma sample values, wherein the another part of the available template area is different from the part of the available template area mentioned in Step 3;
  • Step 6 accumulating, for each of the M filter shapes respectively, the errors between the predicted chroma sample values and the known chroma sample values in the another part of the available template area by Sum of Absolute Difference (SAD), Sum of Squared Difference (SSD), or Sum of Absolute Transformed Difference (SATD), denoted as E o , E b ... E M.b
  • Step 7 sorting and selecting K smallest errors, denoted as E’ o , E’ ... E’ K -I, which correspond to K filter shapes and K filter coefficient sets;
  • Step 8 selecting one filter shape candidate from the K filter shape candidates, to apply to the current CU for chroma prediction. If K is more than 1, then the decoder may still receive signal from the encoder indicating the applied filter. However, if K is 1, then the signaling can be omitted and the filter with the smallest accumulated error is determined to be the applied filter.
  • Figure 20A and Figure 20B illustrate examples of 2-fold training for implicitly filter shape derivation.
  • Figure 20A shows that in template area, even-row region RQ (yellow) is used for training/deriving 4 filter coefficient sets, and odd-row region Ri (red) is used for validation/comparing and sorting costs of 4 filter coefficient sets;
  • Figure 20B shows that in template area, R o (yellow) and Ri (red) are interleaved. It should be understood that RQ and Ri can be exchanged in these examples.
  • one of 4 filter shape candidates is to be selected as the applied filter, while the L-shaped template area is divided into even-numbered and odd-numbered rows or columns.
  • the steps include: predefining 4 filter shape candidates for the current CU (for example, 4 filter shape candidates from Figure 16); dividing the available L-shaped template area (for example, 6 chroma rows and columns for CCCM, note that in CCCM design, each chroma sample involves 6 luma samples for down-sampling) into 2 regions denoted as R o , R wherein, for example, R o is composed of the even rows or columns, and Ri is composed of the odd rows or columns (Figure 20A shows the example that in template area, even-row region RQ is used for training/deriving 4 filter coefficient sets, and odd-row region Ri is used for validation/comparing and sorting costs of 4
  • SUBSTITUTE SHEET (RULE 26) filter coefficient sets); applying 4 filter shape candidates independently to a part of the available template area (for example, region R o ); deriving 4 filter coefficient sets for the 4 filter shapes respectively, denoted as F o , F b ... F 3 ; applying the derived F o , F b ... F 3 filter coefficient sets to the other part of the available template area (for example, region R.
  • the steps include: determining 4 filter shape candidates for the current CU; dividing the available L-shaped template area (for example, 6 chroma rows or columns for CCCM) into 2 regions denoted as Ry, R b wherein the luma samples in Ry and Ri are for example interleaved as shown in the following tables: or and Figure 20B shows the example that in template area, Ro and Ri are interleaved; applying 4 filter shape candidates independently to a part of the available template area (for example, region R o ); deriving 4 filter coefficient sets for the 4 filter shapes respectively, denoted as F o , F b ...
  • SUBSTITUTE SHEET ( RULE 26) available template area (for example, region Ri) by SAD, SSD, or SATD, denoted as E o , E 1
  • E 3 sorting and selecting 2 smallest accumulated errors in E o , E 1 . .. E 3 , denoted as E’ o , E’ 1 which correspond to 2 filter shapes and 2 filter coefficient sets, and based on the received signal from the encoder, selecting one filter shape candidate in the 2 filter shape candidates, to apply to the current CU for chroma prediction.
  • nonlinear term P may also be included in FLM filters (e.g., a 3*2 filter as shown in Figure 11) and derived in a similar way as discussed above.
  • the steps of dividing the available L-shaped template area may be omitted.
  • the M filter coefficient sets may be derived based on the sample values from the available template area and then applied back to the available template area respectively, to predict the corresponding chroma sample values for accumulating the errors.
  • an MLR model (linear equations) must be derived at both the encoder and the decoder.
  • several methods are proposed to derive the pseudo inverse matrix A + , or to directly solve the linear equations.
  • Other known methods like Newton's method, Cayley-Hamilton method, and Eigendecomposition as mentioned in https://en.wikipedia.org/wiki/Invertible_matrix can also be applied.
  • a + can be denoted as A 1 for simplification.
  • the linear equations may be solved as follows
  • A can be firstly decomposed by Cholesky-Crout algorithm, leading to one upper triangular and one lower triangular matrices, and one forward substitution plus one backward substitution can be applied in serial to obtain the solution.
  • Cholesky-Crout algorithm leading to one upper triangular and one lower triangular matrices, and one forward substitution plus one backward substitution can be applied in serial to obtain the solution.
  • 3x3 example shows a 3x3 example.
  • DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblock/Sample levels for example, when predefined l «(bitDepth-l), meanC, meanL, or meanC-meanL (mean current chroma or other chroma, luma values from available, or subset of FLM reconstructed neighboring region).
  • bitDepth-l meanC
  • meanL meanL
  • meanC-meanL mean current chroma or other chroma, luma values from available, or subset of FLM reconstructed neighboring region.
  • REG_SQR can be predefined or signaled or switched in SPS/ DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblock/Sample levels.
  • Figure 11 shows a typical case that the FLM parameters are derived using top 2 and/or left 3 luma lines and top 1 and/or left 1 chroma lines.
  • using different region for parameter derivation may bring coding benefit because of different block content and the reconstructive quality of different neighboring samples, as mentioned above.
  • Several ways to choose the applied region for parameter derivation are proposed below:
  • the FLM derivation can only use top or left luma and/or chroma samples to derive the parameters.
  • FLM L, or FLM_T can be predefined or signaled or switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels.
  • the number of extended luma/chroma samples can be predefined, or signaled or switched in SPS/ DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblock/Sample levels.
  • Figure 13 shows an illustration of FEM E and FLM T (e.g., under 4 tap).
  • FLM L or FLM_T When FLM L or FLM_T is applied, only H’ or W’ luma/chroma samples are used for parameter derivation, respectively.
  • different line index can be predefined, or signaled or switched in SPS/APS/PPS/PH/SH/Region/CTU/CU/Subblock/Sample levels, to indicate the selected luma-chroma sample pair line. This may benefit from different reconstructive quality of different line samples.
  • FIG 14 shows that similar to MRL, FLM can use different lines for parameter derivation (e.g., under 4 tap).
  • FLM can use light blue/yellow luma and/or chroma samples in index 1.
  • the luma sample values of an external region of the video block to be decoded may be referred to as “the external luma sample values”, and the chroma sample values of the external region may be referred to as “the external chroma sample values” throughout the disclosure.
  • Corresponding syntax may be defined as below in Table 9 for the FLM prediction.
  • FLC represents fixed length code
  • TU represents truncated unary code
  • EGk represents exponential-golomb code with order k, where k can be fixed or signaled/switched in SPS/ DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels
  • SVLC represents signed EGO
  • UVLC represents unsigned EGO.
  • a new method for cross-component prediction is proposed on the basis of the existing linear model designs, in order to further improve coding accuracy and efficiency. Main aspects of the proposed method are detailed as follows.
  • reference samples/training template/reconstructed neighboring region usually refers to the luma samples used for deriving the MLR model parameters, which are then applied to the inner luma samples in one CU, to predict the chroma samples in the CU.
  • pre-operations e.g., pre linear weighted, sign, scale/abs, thresholding, ReLU
  • the pre-operations may comprise calculating sample differences based on the luma sample values.
  • the sample differences may be characterized as gradients, and thus this new method is also referred to as gradient linear model (GLM) in certain embodiments.
  • FIGS 15 A to 15D show some examples for l -tap/2-tap (with offset) pre-operations, where 2-tap coefficients are denoted as (a, b).
  • each circle as illustrated in Figures 15A to 15D represent an illustrative chroma position in the YUV 4:2:0 format.
  • a luma sample corresponding to a chroma sample may be obtained by performing a down-sampling operation on more than one (e.g., 4) reconstructed luma samples corresponding to the chroma sample (e.g., located around the chroma sample).
  • the chroma position may correspond to corresponding to one or more luma samples comprising a collocated luma sample.
  • the different 1-tap patterns are designed for different gradient directions and using different “interpolated” luma samples (weighting to different luma location) for gradient calculation. For example, one typical fdter [1 , 0, -1 ; 1, 0, -1 ] is shown in Figures 15 A, 15C and 15D, which represents the following operations:
  • Rec L represents the reconstructed luma sample values and Rec L "(ij) represents the pre-operated luma sample values.
  • the 1-tap filters as shown in Figures 15 A, 15C and 15D may be understood as alternatives for the down-sampling filters as used in CCLM (please refer to equations (6)-(7)), with changed filter coefficients.
  • Pre-operations can be according to gradients, edge direction (detection), pixel intensity, pixel variation, pixel variance, Roberts/Prewitt/compass/Sobel/Laplacian operator, high-pass filter (by calculating gradients or other relevant operators), low-pass filter (by performing weighted-average operations)... etc.
  • the edge direction detectors listed in the examples can be extended to different edge directions. For example, 1-tap (1, -1) or 2-tap (a, b) applied along different directions to detect different edge gradients.
  • the filter shape/coeff can be symmetric with respect to the chroma position, as the Figures 15 A to 15D examples (420 type-0 case).
  • the pre-operation parameters can be fixed or signaled/switched in
  • the pre-operations may relate to calculating sample differences of the luma sample values.
  • the pre-operations may comprise performing downsampling by weighted-average operations.
  • the pre-operations can be applied repeatedly. For example, one may apply one template filtering to template to remove outliers using the low-pass smoothing FIR filter [1, 2, l]/4, or [1, 2, 1; 1, 2, l]/8 (i.e., down-sampling) and then apply 1-tap GLM filter to calculate the sample differences to derive the linear model. It may be contemplated that one may also calculate the sample differences and then enabling down-sampling.
  • the pre-operation coefficients (finally applied (e.g., 3), or middle applied (e g., -1, 4) to per luma sample) can be limited to power -of-2 values to save multipliers.
  • the proposed new method may be reused for/combined with the above discussed CCLM, which utilizing a simple linear regression (SLR) model and using one corresponding luma sample value to predict the chroma sample value. This is also referred to as a 1-tap case.
  • SLR simple linear regression
  • deriving the linear model further comprises deriving a scale parameter a and an offset parameter [3 by using the pre-operated neighboring luma sample values and the neighboring chroma sample values.
  • the linear model may be re-writen as:
  • L here represents “pre-operated” luma samples.
  • the parameter derivation of 1-tap GLM can reuse CCLM design, but taking directional gradient into consideration (may be with high-pass filter).
  • the scale parameter a may be derived by utilizing a division look-up table, as detailed below, to enable simplification.
  • the scale parameter a and the offset paremeter ⁇ may be derived by utilizing the above-discussed min-max method.
  • the scale parameter a and the offset paremeter ⁇ may be derived by: comparing the pre-operated neighboring luma sample values to determine a minimum luma samle value Y A and a maximum luma sample value Y B ; determining corresponding choma samples values X A and X B for the minimum luma samle value Y A and the maximum luma sample value Y B , respectively; and deriving the scale parameter a and the offset paremeter P based on the minimum luma samle value Y A , the maximum luma sample value Y B , and the corresponding choma samples values X A andX B according to the following equations:
  • the encoder may determine a scale adjustment value (for example, “u”) to be signaled in the bitstream and add the scale adjustment value to the derived scale parameter a .
  • the decoder may dertermine the scale adjustment value (for example, “u”) from the bitstream and add the scale adjustment value to the derived scale parameter a .
  • the added value are finally used to predict the internal chroma sample values.
  • the proposed new method may be reused for/combined with FLM, which utilizing a multiple linear regression (MLR) model and using multiple luma sample values to predict the chroma sample value.
  • MLR multiple linear regression
  • This is also referred to as a multi-tap case, for example, 2-tap.
  • the linear model may be re-writen as:
  • multiple scale parameters a and an offset parameter p may be derived by using the pre-operated neighboring luma sample values and the neighboring chroma sample values.
  • the offset parameter p is optional.
  • at least one of the multiple scale parameters a may be derived by utilizing the sample differences.
  • another of the multiple scale parameters a may be derived by utilizing the down- sampled luma sample value.
  • at least one of the multiple scale parameters a may be derived by utilizing horizontal or vertical sample differences calculated on the basis of down-sampled neighboring luma sample values.
  • the linear model may combine multiple scale parameters a asscosicated with different pre-opertaions.
  • the used direction oriented filter shape can be derived at decoder to save bit overhead. For example, at the decoder, a number of directional gradient filters may be applied for each reconstructed luma sample of the L-shaped template of the i-th neighboring row and column of the current block. Then the filtered values (gradients) may be accumulated for each direction of the number of directional gradient filters respectively. In an example, the accumulated value is an accumulated value of absolute values of corresponding filtered values. After the accumulation, the direction of the directional gradient filter for which the accumulated value is the largest may be determined as the derived (luma) gradient direction. For example, a Histogram of Gradients (HoG) may be built to determine the largest value. The derived direction can be further applied as the direction for predicting chroma samples in the current block.
  • HoG Histogram of Gradients
  • DIMD decoder-side intra mode derivation
  • Step 1 applying 2 kinds of directional gradient filters (3x3 hor/ver Sobel) for each reconstructed luma sample of the L-shaped template of the 2 nd neighboring row and column of the current block;
  • Step 2 accumulating filtered values (gradients) by SAD (sum of absolute differences) for each of the directional gradient filters;
  • Step 3 Build a Histogram of Gradients (HoG) based on the accumulating filtered values
  • Step 4 The largest value in HoG is determined to be the derived (luma) gradient direction, based on which the GLM filter may be determined.
  • shape candidates are [-1, 0, 1; -1, 0, 1] (horizontal) and [1, 2, 1 ; -1 , -2, -1 ] (vertical)
  • shape [-1, 0, 1; -1, 0, 1] for GLM based chroma prediction.
  • the gradient filter used for deriving the gradient direction can be the same or different with the GLM filter in shape.
  • both of the filters may be horizontal [-1, 0, 1; -1, 0, 1], or the two filters may have different shapes, while the GLM filter may be determined based on the gradient filter.
  • the proposed GLM can be combined with above discussed MMLM or ELM.
  • each group can share or have its own filter shape, with syntaxes indicating shape for each group.
  • horiontal grandients grad hor may be classified into a first group, which correspond to a first linear model
  • vertical grandients grad_ver may be classified into a second group, which correspond to a second linear model.
  • the horiontal luma patterns may be generated only once.
  • Further possible classifiers are also provided as follows.
  • the neighboring and internal luma-chroma sample pairs of the current video block may be classified into muitple groups based on one or more thresholds
  • each neighboring/internal chroma sample and its corresponding luma sample may be referred to as a luma-chroma sample pair.
  • the one or more thresholds are associated with intensities of neighboring/ internal luma samples.
  • each of the multiple groups corresponds to a respective one of the plurality of linear models.
  • the following operations may be performed: classifying neighboring reconstructed luma-chroma sample pairs of the current video block into 2 groups based on Threshold, deriving different linear models for different groups, wherein the deriving process may be GLM simplified, i.e., with the above preoperations to reduce the number of taps; classifying luma-chroma sample pairs inside the CU (internal luma-chroma sample pairs, wherein each of the internal luma-chroma sample pairs comprises an internal chroma sample value to be predicted with the derived linear model) into 2 groups similarly based on Threshold; applying different linear models to the reconstructed luma samples in different groups; and predicting chroma samples in the CU based on different classified linear models.
  • Threshold may be average value of the neighboring reconstructed luma samples.
  • the number of classes (2) can be extended to multiple classes by increasing the number of Threshold (e.g., equally divided based on min/max of neighboring reconstructed (down-sampled) luma samples, fixed or signaled/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels).
  • the filtered values of FLM/GLM apply on neighboring luma samples are used for classification. For example, if 1- tap (1, -1) GLM is applied, average AC values are used (physical meaning).
  • the processing can be: classifying neighboring reconstructed luma-chroma sample pairs into K groups based on one or more filter shapes, one or more filtered values, and K-l Threshold Ti; deriving different MLR models for different groups, wherein the deriving process may be GLM simplified, i.e., with the above pre-operations to reduce the number of taps, classifying lumachroma sample pairs inside the CU (internal luma-chroma sample pairs, wherein each of the internal luma-chroma sample pairs comprises an internal chroma sample value to be predicted with the derived linear model) into K groups similarly based on one or more filter shapes, one or more filtered values, and K-l Threshold Ti; applying different linear models to the reconstructed luma samples in different groups; predicting chroma samples in the CU based on different classified linear models.
  • Threshold can be predefined (e.g., 0, or can be a table) or signaled/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels).
  • Threshold can be the average AC value (filtered value) (2 groups), or equally divided based on min/max AC (K groups), of neighboring reconstructed (can be down- sampled) luma samples.
  • one filter shape (e.g., 1-tap) may be selected to calculate edge strengths.
  • the direction is determined as a direction along which a sample difference between samples of the current and N neighboring samples (e.g., all 6 luma samples) is calculated.
  • the filter shape [1, 0, -1; 1, 0, -1]
  • the filter at the upper middle in Figure 15 A indicates a horizontal direction since a sample difference may be calculated between samples in the horizontal direction
  • the filter below it shape [1, 2, 1; -1, -2, -1] indicates a vertical direction since a sample difference may be calculated between samples in the vertical direction.
  • the positive and negative coefficients in each of the filters enable the calculation of the sample differences.
  • the filter shape used for classification can be the same or different with the filter shape used for MLR prediction.
  • Both and the number of thresholds M-l, the thresholds values Ti can be fixed or signaled/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels.
  • other classifiers/combined-classifiers as discussed in ELM can also be used for GLM.
  • the matrix/parameter derivation in FLM requires floating-point operation (e g., division in closed-form), which is expensive for decoder hardware, so a fixed-point design is required.
  • floating-point operation e g., division in closed-form
  • CCLM modified luma reconstructed sample generation of CCLM
  • the original CCLM process can be reused for GLM, including fixed-point operation, MDLM downsampling, division table, applied size restriction, min-max approximation, and scale adjustment.
  • 1-tap GLM can have its own configurations or share the same design as CCLM.
  • each group can apply the same or different simplification operation. For example, samples for each group are padded respectively to the target sample number before applying right shift, and then apply the same derivation process, same division table.
  • the 1-tap case can reuse the CCLM design, dividing by n may be implemented by right shift, dividing by A 2 may be implemented by by a LUT.
  • the integerization parameters including n ⁇ , n A1 , n A2 , r A1 , r A2 n table invloved in the integerization design of LMS CCLM and intermediate parameters for deriving the linear model (equations (19)-(20)) can be the same as CCLM or have different values, to have more precision.
  • the padding method for GLM can be the same or different with that of CCLM.
  • Division LUT proposed for CCLM/LIC Long Illumination Compensation
  • AVC/HEVC/AV1/VVC/AVS can be used for GLM division.
  • CCLM/LIC Large Illumination Compensation
  • the division LUT can be different from CCLM.
  • CCLM uses min-max with DivTable as in equation 5, but GLM uses 32-entries LMS division LUT as in Table 5.
  • the meanL values may not always be positive (e.g., using filtered/gradient values to classify groups), so sgn(meanL) needs to be extracted, and use abs(meanL) to look-up the division LUT.
  • division LUT used for MMLM classification and parameter derivation can be different. For example, using lower precision LUT (as the LUT in min-max) for mean classification, and using higher precision LUT (as in the LMS) for parameter derivation.
  • ELM/FLM/GLM Similar to the CCLM design, some size restrictions can be applied for ELM/FLM/GLM. For example, same constraint for luma-chroma latency in dual tree may be applied.
  • the size restriction can be according to the CU area/width/height/depth.
  • the threshold can be predefined or signaled in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels.
  • the predefined threshold may be 128 for chroma CU area.
  • the at least one pre-operation is performed in response to determining that the video block meets an enabling threshold, wherein the enabling threshold is associated with area, width, height or partition depth of the video block. Spesifically, the enabling threshold may define a minium or maximum area, width, height or partition depth of the video block.
  • the video block may comprise a current chroma block and its collocated luma block. It is also proposed to apply the above enabling threshold for the current chroma block and its collocated luma block jointly. For example, the at least one pre-operation is performed in response to determining the enabling threshold is met for both the current chroma block and its collocated luma block.
  • the top template samples generation can be limited to 1 row, to reduce CTU row line buffer storage. Note that only one luma line (general line buffer in intra prediction) is used to make the down-sampled luma samples when the upper reference line is at the CTU boundary.
  • top template can be limited to only use 1 row (but not 2) for parameter derivation (other CUs can still use 2 rows). This saves luma sample line buffer storage when processing CTU row by row at decoder hardware.
  • Several methods can be used to achieve the line buffer reduction. Note the example of limited “1” row can be extended to N rows with similar operations. Similarly, 2-tap or multi-tap can also apply such operations. When applying multi-tap, chroma samples may also need to apply operations.
  • [00235] For example, take the 1 -tap filter [1 , 0, -1 ; 1, 0, -1 ] shown in Figure 15A as an example for illustration.
  • This filter can be reduced to [0, 0, 0; 1, 0, -1], i.e., only use below row coefficients.
  • the limited upper row luma samples can be padded (repetitive, mirror, 0, meanL, meanC. . .etc.) from the bellow row luma samples.
  • T ake an example where N 4, that is, the video block is at a top boundary of a current CTU, while top 4 rows of neighboring luma sample values and corresponding chroma sample values are used for deriving the linear model.
  • the corresponding chroma sample values may refer to corresponding top 4 rows of neighboring chroma sample values (for example, for the YUV 4:4:4 format).
  • the corresponding chroma sample values may refer to corresponding top 2 rows of neighboring chroma sample values (for example, for the YUV 4:2:0 format).
  • the top 4 rows of neighboring luma sample values and corresponding chroma sample values may be divided into two regions - a first region comprising valid sample values (for example, the one nearest row of luma sample values and corresponding chroma sample values) and a second region comprising invalid sample values (for example, the other three rows of luma sample values and corresponding chroma sample values). Then coefficients of the filter corresponding to sample positions not belonging to the first region may be set as zeros, such that only sample values from the first region are used for calculating the sample differences.
  • the filter [1, 0, -1 ; 1 , 0, -1 ] can be reduced to [0, 0, 0; 1 , 0, -1 ].
  • the nearest sample values in the first region may be padded to the second region, such that the padded sample values may be used to calculate the sample differences.
  • pred0 is the predictor based on non-LM mode
  • pred1 is the predictor based on GLM
  • pred0 is the predictor based on one of CCLM (including all MDLM/MMLM)
  • pred1 is the predictor based on GLM
  • pred1 is the predictor based on GLM
  • pred1 is the predictor based on GLM.
  • Different I/P/B slices can have different designs for weights, w0 and wl, depending on if neighboring blocks is coded with CCLM/GLM/other coding mode or the block size/width/height.
  • GLM has good gain complexity trade-off since it can reuse the existing CCLM module without introducing additional derivation.
  • Such 1-tap design can be extended or generalized further according to one or more aspects of the present disclosure.
  • one single corresponding luma sample L may be generated by combining collocated luma sample and neighboring luma samples.
  • the combination may be a combination of different linear filters, e.g., a combination of a high-pass gradient filter (GLM) and a low-pass smoothing filter (e.g., [1, 2, 1, 1, 2, 1 ]/8 FIR down-sampling filter that may be generally used in CCLM); and/or a combination of a linear filter and a non-linear filter (e.g., with power of n, e.g., L n , n can be positive, negative, or +-fractional number (e.g., +1/2, square root or +3, cube, which can rounding and rescale to bitdepth dynamic range)).
  • LLM high-pass gradient filter
  • a low-pass smoothing filter e.g., [1, 2, 1, 1, 2, 1 ]/8 FIR down-sampling filter that may be generally used in CCLM
  • the combination may be repeatedly applied.
  • a combination of GLM and [1, 2, 1; 1, 2, 1]/8 FIR may be applied on the reconstructed luma samples, and then a non-linear power of 1/2 may be applied.
  • the nonlinear filter may provide options when linear filter cannot handle the luma-chroma relationship efficiently. Whether to use nonlinear term can be predefined or signal ed/ switched in SP S/DP S/VP S/ SEI/ AP S/PP S/PH/SH/Regi on/CTU/CU/ Subbl cok/ S ample levels.
  • the GLM may refer to Generalized Linear Model (may be used to generate one single luma sample linearly or nonlinearly, and the generated one single luma sample may be fed into the CCLM linear model to derive parameters of the CCLM linear model), linear/nonlinear generation may be called general patterns. Different gradient or general patterns can be combined to form another pattern.
  • Generalized Linear Model may be used to generate one single luma sample linearly or nonlinearly, and the generated one single luma sample may be fed into the CCLM linear model to derive parameters of the CCLM linear model
  • linear/nonlinear generation may be called general patterns. Different gradient or general patterns can be combined to form another pattern.
  • a gradient pattern may be combined with a CCLM down-sampled value; a gradient pattern may be combined with a non-linear L 2 value; a gradient pattern may be combined with another gradient pattern, the two gradient patterns to be combined may have different directions or the same direction, e.g., [1, 1, 1; -1, -1, -1] and [1, 2, 1; -1, -2, -1], which both have a vertical direction, may be combined, also [1, 1, 1, -1, -1, -1] and [1, 0, -1; 1, 0, -1], which have a vertical and horizontal directions, may be combined, as shown in Figures 15A to 15D.
  • the combination may comprise plus, minus, or linear weighted.
  • pre-operations can be applied repeatedly and GLM can be applied on pre linear weighted/pre-operated samples.
  • CCLM one template filtering can be applied to luma samples, in order to remove outliers using the low-pass smoothing FIR fdter [1, 2, 1; 1, 2, l]/8 (i.e., CCLM down-sampling smoothing filter) and to generate down-sampled luma samples (one down-sampled luma sample corresponding to one chroma sample).
  • 1-tap GLM can be applied on smoothed down-sampled luma samples to derive the MLR model.
  • Some gradient filter patterns such as 3x3 Sobel or Prewitt operators, can be applied on down-sampled luma samples.
  • the following table shows some of the gradient filter patterns.
  • the gradient filter patterns can be combined with other gradient/general filter patterns in the down-sampled luma domain.
  • a combined filter pattern may be applied on down-sampled luma samples.
  • the combined filter pattern may be derived by performing addition or subtraction operations to respective coefficients of the gradient filter pattern and a DC/low-pass based filter pattern, such as filter pattern [0, 0, 0; 0, 1, 0; 0, 0, 0], or [1, 2, 1; 2, 4, 1; 1, 2, 1],
  • the combined filter pattern is derived by performing addition or subtraction operations to a coefficient of the gradient filter pattern and a non-linear value such as L 2 .
  • the combined filter pattern is derived by performing addition or subtraction operations to respective coefficients of the gradient filter pattern and another gradient filter pattern having a different or the same direction.
  • the combined filter pattern is derived by performing linear weighted operations to the coefficients of the gradient filter pattern.
  • GLM applied on down-sampled domain can fit in CCCM framework but may sacrifice high frequency accuracy since low-pass smoothing is applied before applying GLM.
  • GLM serves as input of CCCM
  • CCCM applies luma down-sampling before convolution as with CCLM.
  • the reconstructed luma samples are down-sampled to match the lower resolution chroma grid when chroma sub-sampling is used.
  • 1-tap GLM can also be taken as changing CCLM down-sample filter coefficients (e.g., from [1, 2, 1; 1, 2, 1]/ 8 to [1, 2, 1, -1, -2, -1], i.e., from low-pass to high-pass), the GLM can serve as the input of CCCM.
  • the gradient filter of GLM replaces luma down-sampling filter ([1, 2, 1, 1, 2, 1]/ 8) with gradient-based coefficients (e.g., [1, 2, 1, -1, -2, -1).
  • CCCM operation becomes “linear/non-linear combination of gradients”, as shown by the following equation:
  • predChromaVal c 0 C + c 1 N + c 2 S + c 3 E + c 4 W + c 5 P + c 6 B
  • C, N, S, E, W, P are gradients of current or neighboring samples (compared to original down-sample values for CCCM).
  • Related GLM methods described in this disclosure can be applied in the same way before entering CCCM convolution, e.g., classification, separate Cb/Cr control, syntax, pattern combining, PU size restriction etc.
  • the gradient-based coefficients replacement can apply to specific CCCM taps. Also, not only high-pass but low-pass/band-pass/all-pass coefficients replacement can be used. The replacement can be combined with FLM/CCCM shape switch discussed above (leading to different number of taps). For example, gradient patterns in Figures 15A to 15D can be used for replacement.
  • the operations for applying GLM as input of CCCM includes: predefining one or more coefficients candidates for CCCM/FLM down-sampling; determining CCCM/FLM filter shape and number of filter taps for this CU; applying different CCLM downsample coefficients to different filter taps, wherein the coefficients can be high-pass filters (GLM), or low-pass/band-pass/all-pass filters; generating the down-sampled luma samples (using the applied coefficients) for CCCM input samples; and feeding the generated down- sampled luma samples into CCCM process.
  • GLM high-pass filters
  • the coefficients candidate for CCCM/FLM downsampling can be predefined or signaled/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels.
  • one or more syntaxes may be introduced to indicate information on the GLM.
  • GLM syntaxes is illustrated in the following Table 10.
  • EGk exponential-golomb code with order k, where k can be fixed or signaled/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblcok/Sample levels.
  • the GLM on/off control for Cb/Cr components may be jointly or separately. For example, at CU level, 1 flag may be used to indicate if GLM is active for this CU. If active, 1 flag may be used to indicate if Cb/Cr are both active. If not both active, 1 flag to indicate either Cb or Cr is active. Filter index/gradient (general) pattern may be signaled separately when Cb and/or Cr is active. All flags may have its own context model or be bypass coded. [00271] In another aspect of the present disclosure, whether to signal GLM on/off flags may depend on luma/chroma coding modes, and/or CU size.
  • GLM may be inferred off when MMLM or MMLM L or MMLM_T is applied; when CU area ⁇ A, where A can be predefined or signal ed/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblock/Sample levels; If combined with CCCM, GLM may be inferred off when CCCM is on.
  • CCCM requires to process down-sampled luma reference values before the calculation of model parameters and applying the CCCM model, which burden decoder processing cycles.
  • CCCM without down-sampling process is proposed, including utilizing non-down-sampled luma reference values and/or different selection of nondown-sampled luma reference.
  • One or more filter shapes may be used for the purpose as described below.
  • reference samples/training templates/reconstructed neighboring regions usually refer to the luma samples used for deriving the MLR model parameters, which are then applied to the inner luma samples in one CU, to predict the chroma samples in the CU.
  • Filter shape
  • One or more shape/number of filter taps may be used for CCCM prediction, as shown in Figure 16, Figure 17, and Figures 18A to 18B.
  • One or more sets of filter taps may be used for FLM prediction, examples being shown in Figures 19A to 19G.
  • the selected luma reference values are non-down-sampled.
  • One or more predefined shape/number of filter taps may be used for CCCM prediction based on previous decoded information on TB/CB/slice/picture/sequence level.
  • a multiple tap filter can fit well on training data (i.e., top/left neighboring reconstructed luma/chroma samples), in some cases, that training data do not capture full characteristics of the testing data, and it may result in overfitting and may not predict well on the testing data (i.e., the to-be-predicted chroma block samples).
  • different filter shapes may adapt well to different video block content, leading to more accurate prediction.
  • the filter shape/number of filter taps can be predefined or signaled/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblock/Sample levels.
  • a set of filter shape candidates can be predefined or signaled/switched in SPS/DPS/VPS/SEI/APS/PPS/PH/SH/Region/CTU/CU/Subblock/Sample levels.
  • Different components U/V may have different filter switch control.
  • the filter shape selection of U/V components can be switched in PH or in CU/CTU levels.
  • N-tap can represent N-tap with or without the offset ⁇ as described above.
  • Different chroma types/color formats can have different predefined filter shapes/taps. For example, using predefined filter shape for 420 type-0: (1, 2, 4, 5), 420 type-2: (0, 1, 2, 4, 7), 422: (1, 4), 444: (0, 1, 2, 3, 4, 5) as shown in Figure 12.
  • the unavailable luma/chroma samples for deriving the MLR model can be padded from available reconstructed samples. For example, if using a 6-tap (0, 1, 2, 3, 4, 5) filter as in Figure 12, for a CU located at the left picture boundary, the left columns including (0, 3) are not available (out of picture boundary), so (0, 3) are repetitive padding from (1, 4) to apply the 6-tap filter. Note the padding process applied in both training data (top/left neighboring reconstructed luma/chroma samples) and testing data (the luma/chroma samples in the CU).
  • the unavailable luma/chroma samples for deriving the MLR model can be skipped and not used. Then the padding process is not needed for the unavailable luma/chroma samples.
  • CCCM requires to process LDL decomposition to calculate the model parameters of CCCM model, avoiding using square root operations and only integer arithmetic is required.
  • CCLM/MMLM with LDL decomposition are proposed.
  • LDL decomposition may also be used in ELM/FLM/GLM, as described above.
  • reference samples/training templates/reconstructed neighboring regions usually refer to the luma samples used for deriving the MLR model parameters, which are then applied to the inner luma samples in one CU, to predict the chroma samples in the CU.
  • One or more reference samples may be used for CCLM/MMLM prediction, i.e., as shown in Figure 10B, the reference area may be the same as the reference area in CCCM. Different reference areas may be used for CCLM/MMLM prediction based on previous decoded information on TB/CB/slice/picture/sequence level.
  • training data with multiple reference areas can fit well on the calculation of model parameters, in some cases that training data do not capture full characteristics of testing data, it may result in overfitting and may not predict well on testing data (i.e., the to-be- predicted chroma block samples). Also, different reference areas may adapt well to different video block content, leading to more accurate prediction. To address this issue, the reference shape/number of reference areas can be predefined or signal ed/switched in
  • a set of reference area candidates can be predefined or signaled/switched in
  • the unavailable luma/chroma samples for deriving the MLR model can be padded from available reconstructed samples, the padding process being applied in both training data (top/left neighboring reconstructed luma/chroma samples) and testing data (the luma/chroma samples in the CU).
  • the unavailable luma/chroma samples for deriving the MLR model can be skipped and not used. Then the padding process is not needed for the unavailable luma/chroma samples.
  • FLM requires to process down-sampled luma reference values and calculate model parameters, which burden decoder processing cycles, especially for small blocks.
  • FLM with minimal samples restriction is proposed, for example, FLM is only used for samples larger than predefined number, such as 64, 128.
  • predefined number such as 64, 128.
  • FLM is only used in single model for samples larger than a predefined number, such as 256, and FLM is only used in multi model for samples larger than a predefined number, such as 128.
  • the number of predefined minimal samples for single model may be larger than or equal to the number of predefined minimal samples for multi model.
  • FLM/GLM/ELM/CCCM is only used in single model for samples larger than or equal to a predefined number, such as 128, and FLM/GLM/ELM/CCCM is only used in multi model for samples larger than or equal to a predefined number, such as 256.
  • the number of predefined minimal samples for FLM/GLM/ELM may be larger than or equal to the number of predefined minimal samples for CCCM.
  • CCCM is only used in single model for samples larger than or equal to a predefined number, such as 0, and CCCM is only used in multi model for samples larger than or equal to a predefined number, such as 128.
  • FLM is only used in single model for samples larger than or equal to a predefined number, such as 128, and FLM is only used in multi model for samples larger than or equal to a predefined number, such as 256.
  • Figure 21 illustrates a workflow of a method 2100 for decoding video data according to one or more aspects of the present disclosure.
  • the method 2100 comprises obtaining a video block from a bitstream.
  • the method 2100 comprises obtaining internal luma sample values of the video block, external luma sample values of an external region of the video block and external chroma sample values of the external region.
  • the method 2100 comprises determining, based on the external luma sample values and the external chroma sample values, a set of weighting coefficients corresponding to a filter shape, wherein the filter shape and the set of weighting coefficients are configured for predicting an chroma sample value based on a plurality of corresponding luma sample values, and wherein the set of weighting coefficients includes a weighting coefficient for a non-linear value derived by scaling the product of two of a plurality of corresponding luma sample values or a square of one of the plurality of corresponding luma sample values to a luma sample value range of the video block.
  • the method 2100 comprises predicting, with the filter shape and the set of weighting coefficients, the internal chroma sample values based on the internal luma sample values.
  • the method 2100 comprises obtaining decoded video block using the predicted internal chroma sample values.
  • the set of weighting coefficients further includes one or more weighting coefficients for one or more luma sample values of the plurality of corresponding luma sample values respectively.
  • the set of weighting coefficients further includes a weighting coefficient for a bias value.
  • the plurality of corresponding luma sample values comprise a plurality of pre-operated luma sample values obtained by weighted-average operations.
  • the plurality of corresponding luma sample values comprise a first pre-operated luma sample value collocated with the chroma sample value to be predicted, a second pre-operated luma sample value left to the first pre-operated luma sample value, a third pre-operated luma sample value right to the first pre-operated luma sample value, a fourth preoperated luma sample value above the first pre-operated luma sample value, and a fifth preoperated luma sample value below the first pre-operated luma sample value.
  • the product of two of the plurality of corresponding luma sample values comprises a product of any two of the first pre-operated luma sample value, the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth preoperated luma sample value, and the fifth pre-operated luma sample value; and the square of one of the plurality of corresponding luma sample values comprises a square of one of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value.
  • the plurality of corresponding luma sample values further comprise an average value of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value
  • the square of one of the plurality of corresponding luma sample values comprises a square of the average value of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value.
  • the plurality of corresponding luma sample values further comprise an average value of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value
  • the product of two of the plurality of corresponding luma sample values comprises a product of the first pre-operated luma sample value and the average value of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value.
  • the plurality of corresponding luma sample values comprise a first luma sample value collocated with the chroma sample value to be predicted, a second luma sample value left to the first luma sample value, a third luma sample value right to the first luma sample value, a fourth luma sample value below the first luma sample value, a fifth luma sample value below the second luma sample value, and a sixth luma sample value below the third luma sample value.
  • the product of two of the plurality of corresponding luma sample values comprises a product of any two of the first luma sample value, the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value; and the square of one of the plurality of corresponding luma sample values comprises a square of one of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value or the sixth luma sample value.
  • the plurality of corresponding luma sample values further comprise an average value of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value
  • the square of one of the plurality of corresponding luma sample values comprises a square of the average value of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value.
  • the plurality of corresponding luma sample values further comprise an average value of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value
  • the product of two of the plurality of corresponding luma sample values comprises a product of the first luma sample value and the average value of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value.
  • Figure 22 illustrates a workflow of a method 2200 for encoding video data according to one or more aspects of the present disclosure.
  • the method 2200 comprises obtaining a video block.
  • the method 2200 comprises obtaining internal luma sample values of the video block, external luma sample values of an external region of the video block and external chroma sample values of the external region.
  • the method 2200 comprises determining, based on the external luma sample values and the external chroma sample values, a set of weighting coefficients corresponding to a filter shape, wherein the filter shape and the set of weighting coefficients are configured for predicting an chroma sample value based on a plurality of corresponding luma sample values, and wherein the set of weighting coefficients includes a weighting coefficient for a non-linear value derived by scaling the product of two of a plurality of corresponding luma sample values or a square of one of the plurality of corresponding luma sample values to a luma sample value range of the video block.
  • the method 2200 comprises predicting, with the filter shape and the set of weighting coefficients, the internal chroma sample values based on the internal luma sample values.
  • the method 2200 comprises generating a bitstream comprising encoded video block by using the predicted internal chroma sample values.
  • the set of weighting coefficients further includes one or more weighting coefficients for one or more luma sample values of the plurality of corresponding luma sample values respectively. [00322] In one example, the set of weighting coefficients further includes a weighting coefficient for a bias value.
  • the plurality of corresponding luma sample values comprise a plurality of pre-operated luma sample values obtained by weighted-average operations.
  • the plurality of corresponding luma sample values comprise a first pre-operated luma sample value collocated with the chroma sample value to be predicted, a second pre-operated luma sample value left to the first pre-operated luma sample value, a third pre-operated luma sample value right to the first pre-operated luma sample value, a fourth preoperated luma sample value above the first pre-operated luma sample value, and a fifth preoperated luma sample value below the first pre-operated luma sample value.
  • the product of two of the plurality of corresponding luma sample values comprises a product of any two of the first pre-operated luma sample value, the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth preoperated luma sample value, and the fifth pre-operated luma sample value; and the square of one of the plurality of corresponding luma sample values comprises a square of one of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value.
  • the plurality of corresponding luma sample values further comprise an average value of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value
  • the square of one of the plurality of corresponding luma sample values comprises a square of the average value of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value.
  • the plurality of corresponding luma sample values further comprise an average value of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value
  • the product of two of the plurality of corresponding luma sample values comprises a product of the first pre-operated luma sample value and the average value of the second pre-operated luma sample value, the third pre-operated luma sample value, the fourth pre-operated luma sample value, and the fifth pre-operated luma sample value.
  • the plurality of corresponding luma sample values comprise a first luma sample value collocated with the chroma sample value to be predicted, a second luma sample value left to the first luma sample value, a third luma sample value right to the first luma sample value, a fourth luma sample value below the first luma sample value, a fifth luma sample value below the second luma sample value, and a sixth luma sample value below the third luma sample value.
  • the product of two of the plurality of corresponding luma sample values comprises a product of any two of the first luma sample value, the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value; and the square of one of the plurality of corresponding luma sample values comprises a square of one of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value or the sixth luma sample value.
  • the plurality of corresponding luma sample values further comprise an average value of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value
  • the square of one of the plurality of corresponding luma sample values comprises a square of the average value of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value.
  • the plurality of corresponding luma sample values further comprise an average value of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value
  • the product of two of the plurality of corresponding luma sample values comprises a product of the first luma sample value and the average value of the second luma sample value, the third luma sample value, the fourth luma sample value, the fifth luma sample value, and the sixth luma sample value.
  • FIG. 23 illustrates an exemplary computing system 2300 according to one or more aspects of the present disclosure.
  • the computing system 2300 may comprise at least one processor 2310.
  • the computing system 2300 may further comprise at least one storage device 2320.
  • the storage device 2320 may store computer-executable instructions that, when executed, cause the processor 2310 to perform the steps of methods described above.
  • the processor 2310 may be a general-purpose processor, or may also be implemented as a combination of computing devices, e g., a combination of a DSP and a microprocessor, multiple microprocessors, one or more microprocessors in conjunction with a DSP core, or any other such configuration.
  • the storage device 2320 may store the input data, output data, data generated by processor 2310, and/or instructions executed by processor 2310.
  • the storage device 2320 may store computerexecutable instructions that, when executed, cause the processor 2310 to perform any operations according to the embodiments of the present disclosure.
  • the embodiments of the present disclosure may be embodied in a computer- readable medium such as non-transitory computer-readable medium.
  • the non-transitory computer-readable medium may comprise instructions that, when executed, cause one or more processors to perform any operations according to the embodiments of the present disclosure.
  • the instructions, when executed may cause one or more processors to receive a bitstream and perform the decoding operations as described above.
  • the instructions, when executed may cause one or more processors to perform the encoding operations and transmit a bitstream comprising the encoded video information associated with the predicted chroma sample as described above.
  • modules in the methods described above may be implemented in various approaches. These modules may be implemented as hardware, software, or a combination thereof. Moreover, any of these modules may be further functionally divided into sub-modules or combined together.

Landscapes

  • Engineering & Computer Science (AREA)
  • Multimedia (AREA)
  • Signal Processing (AREA)
  • Compression Or Coding Systems Of Tv Signals (AREA)

Abstract

La présente divulgation concerne un procédé permettant de décoder des données vidéo, consistant à obtenir un bloc vidéo à partir d'un flux binaire ; à obtenir des valeurs d'échantillon de luminance interne du bloc vidéo, des valeurs d'échantillon de luminance externe d'une région externe du bloc vidéo et des valeurs d'échantillon de chrominance externe de la région externe ; à déterminer un ensemble de coefficients de pondération correspondant à une forme de filtre, qui sont configurés pour prédire une valeur d'échantillon de chrominance sur la base de valeurs d'échantillon de luminance correspondantes, et l'ensemble de coefficients de pondération comprenant un coefficient de pondération pour une valeur non linéaire dérivée en mettant à l'échelle le produit de deux valeurs d'une pluralité de valeurs d'échantillon de luminance correspondantes ou un carré d'une valeur de la pluralité de valeurs d'échantillon de luminance correspondantes à une plage de valeurs d'échantillon de luminance du bloc vidéo ; à prédire, avec la forme de filtre et l'ensemble de coefficients de pondération, les valeurs d'échantillon de chrominance interne sur la base des valeurs d'échantillon de luminance interne et à obtenir un bloc vidéo décodé.
PCT/US2023/030091 2022-08-12 2023-08-11 Procédé et appareil de prédiction entre composantes pour un codage vidéo WO2024035939A1 (fr)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
US202263397731P 2022-08-12 2022-08-12
US63/397,731 2022-08-12

Publications (1)

Publication Number Publication Date
WO2024035939A1 true WO2024035939A1 (fr) 2024-02-15

Family

ID=89852418

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2023/030091 WO2024035939A1 (fr) 2022-08-12 2023-08-11 Procédé et appareil de prédiction entre composantes pour un codage vidéo

Country Status (1)

Country Link
WO (1) WO2024035939A1 (fr)

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170332098A1 (en) * 2016-05-16 2017-11-16 Qualcomm Incorporated Loop sample processing for high dynamic range and wide color gamut video coding
US20200007882A1 (en) * 2017-08-22 2020-01-02 Panasonic Intellectual Property Corporation Of America Image encoder, image decoder, image encoding method, and image decoding method
US20200137394A1 (en) * 2018-10-26 2020-04-30 Mediatek Inc. Luma-based chroma intra-prediction method that utilizes down-sampled luma samples derived from weighting and associated luma-based chroma intra-prediction apparatus

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20170332098A1 (en) * 2016-05-16 2017-11-16 Qualcomm Incorporated Loop sample processing for high dynamic range and wide color gamut video coding
US20200007882A1 (en) * 2017-08-22 2020-01-02 Panasonic Intellectual Property Corporation Of America Image encoder, image decoder, image encoding method, and image decoding method
US20200137394A1 (en) * 2018-10-26 2020-04-30 Mediatek Inc. Luma-based chroma intra-prediction method that utilizes down-sampled luma samples derived from weighting and associated luma-based chroma intra-prediction apparatus

Similar Documents

Publication Publication Date Title
JP5606625B2 (ja) ビデオ符号化のための高度な動きモデルを使った参照処理
CN110830797B (zh) 视频解码装置、视频编码装置和传输比特流的方法
WO2017084577A1 (fr) Procédé et appareil pour un mode d'intraprédiction utilisant un filtre d'intraprédiction en compression vidéo et image
KR102359415B1 (ko) 비디오 코딩을 위한 인터 예측 장치 및 방법을 위한 보간 필터
US10412402B2 (en) Method and apparatus of intra prediction in video coding
WO2023225013A1 (fr) Prédiction de composante transversale améliorée pour codage vidéo
WO2024076632A1 (fr) Prédiction de valeur de coefficient et définition de coût
WO2024035939A1 (fr) Procédé et appareil de prédiction entre composantes pour un codage vidéo
US11202082B2 (en) Image processing apparatus and method
WO2024006409A1 (fr) Procédé et appareil de prédiction inter-composantes pour codage vidéo
WO2024072945A1 (fr) Procédé et appareil de prédiction inter-composantes pour codage vidéo
WO2023250115A1 (fr) Procédé et appareil de prédiction inter-composantes pour codage vidéo
WO2024026098A1 (fr) Procédé et appareil de prédiction inter-composantes pour codage vidéo
WO2023200966A1 (fr) Procédé et appareil de prédiction inter-composantes pour codage vidéo
WO2023230152A1 (fr) Procédé et appareil de prédiction inter-composantes pour codage vidéo
WO2023183510A1 (fr) Procédé et appareil de prédiction de composantes croisées pour un codage vidéo
WO2024107967A2 (fr) Procédé et appareil de prédiction inter-composantes pour un codage vidéo
WO2024130123A2 (fr) Procédé et appareil de prédiction inter-composantes pour codage vidéo
WO2024124188A1 (fr) Procédé et appareil de prédiction inter-composantes pour codage vidéo
WO2024148048A2 (fr) Procédé et appareil de prédiction inter-composantes pour codage vidéo
WO2024081291A1 (fr) Procédé et appareil de prédiction inter-composantes pour un codage vidéo
WO2023154410A1 (fr) Procédé et appareil de pprédiction inter-composante pour codage vidéo
WO2023239676A1 (fr) Prédiction inter-composant améliorée pour un codage vidéo
CN117596393A (zh) 视频解码或编码方法、电子设备、存储介质及程序产品
WO2023249901A1 (fr) Prédiction inter-composantes améliorée pour codage vidéo

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 23853397

Country of ref document: EP

Kind code of ref document: A1