Academia.eduAcademia.edu

Figure 8 – uploaded by junaid Mir

See full PDFdownloadDownload figure
Fig.8 Quality versus rate performance of the evaluated HDR video coding methods using four major HDR quality metrics

Figure 8 Quality versus rate performance of the evaluated HDR video coding methods using four major HDR quality metrics

Related Figures (13)

Fig. 1 High-level diagram of HDR video content workflow
Fig. 1 High-level diagram of HDR video content workflow
Fig. 2 General block diagram of the two-layer backward-compatible HDR video coding methods Fig.3 General block diagram of non-backward-compatible HDR video coding methods
Fig. 2 General block diagram of the two-layer backward-compatible HDR video coding methods Fig.3 General block diagram of non-backward-compatible HDR video coding methods
3.1 HDR-MPEG structed HDR frame from the decoded BL LDR frame. Imperceptible noise is then removed from the residual frame using a visual model, and the residual sequence is encoded in the EL using 8 bits. The key advantage of HDR-MPEG is the freedom of choosing the BL LDR representation of the HDR content. Unlike similar methods [19,20,22], no restric- tion is imposed on the BL LDR content generation which can be produced through a TMO (automatic) or color grad- ing (manual) by a color artist. Implementation details of the method are given in [17]. The filtering operation of the orig- inal method is not implemented for a fair comparison with other HDR video coding methods, and HEVC is used for the encoding of both layers. Mantiuk et al. [17] were the first to propose a two-layer backward-compatible HDR video coding method known as HDR-MPEG. It is a residual-based method in which after conversion of both the HDR and the respective LDR con- tent into a perceptually uniform compatible colorspace, i.e., L’'u'v', areconstruction function (RF) is computed to exploit the correlation between the two representations.
3.1 HDR-MPEG structed HDR frame from the decoded BL LDR frame. Imperceptible noise is then removed from the residual frame using a visual model, and the residual sequence is encoded in the EL using 8 bits. The key advantage of HDR-MPEG is the freedom of choosing the BL LDR representation of the HDR content. Unlike similar methods [19,20,22], no restric- tion is imposed on the BL LDR content generation which can be produced through a TMO (automatic) or color grad- ing (manual) by a color artist. Implementation details of the method are given in [17]. The filtering operation of the orig- inal method is not implemented for a fair comparison with other HDR video coding methods, and HEVC is used for the encoding of both layers. Mantiuk et al. [17] were the first to propose a two-layer backward-compatible HDR video coding method known as HDR-MPEG. It is a residual-based method in which after conversion of both the HDR and the respective LDR con- tent into a perceptually uniform compatible colorspace, i.e., L’'u'v', areconstruction function (RF) is computed to exploit the correlation between the two representations.
where JN is the total number of frames encoded for the test sequence. max(Y;) and min(Y;) are the minimum and maxi- mum luminance values of the j‘“ HDR frame. resnectivelv. The HDR content from BBC and Technicolor is captured using a rig of two Sony F65 cameras simultaneously cap- turing different exposures with neutral density filter. The Technicolor content was initially captured in 4K resolu- tion and then down sampled to 2K resolution. The HDR test sequences from LiU-HDRv repository were captured using the methods and algorithm described in [49]. All these test sequences except from BBC provide HDR frames in OpenEXR half float format with pixel values representing the linear luminance values proportional to the absolute lumi- nance of the scene. Test sequences from BBC were provided in HLG encoded format and were converted back into the linear luminance domain by inverting the HLG curve. Fig- ure 5 shows the example tone mapped frames of these test sequences. 4 Experimental Settings This section provides information regarding the test sequen- ces, the procedures adopted to prepare the HDR content for encoding and the coding conditions used for the experimental performance evaluation.
where JN is the total number of frames encoded for the test sequence. max(Y;) and min(Y;) are the minimum and maxi- mum luminance values of the j‘“ HDR frame. resnectivelv. The HDR content from BBC and Technicolor is captured using a rig of two Sony F65 cameras simultaneously cap- turing different exposures with neutral density filter. The Technicolor content was initially captured in 4K resolu- tion and then down sampled to 2K resolution. The HDR test sequences from LiU-HDRv repository were captured using the methods and algorithm described in [49]. All these test sequences except from BBC provide HDR frames in OpenEXR half float format with pixel values representing the linear luminance values proportional to the absolute lumi- nance of the scene. Test sequences from BBC were provided in HLG encoded format and were converted back into the linear luminance domain by inverting the HLG curve. Fig- ure 5 shows the example tone mapped frames of these test sequences. 4 Experimental Settings This section provides information regarding the test sequen- ces, the procedures adopted to prepare the HDR content for encoding and the coding conditions used for the experimental performance evaluation.
Table 2 Characteristics of HDR test sequences translates the peak luminance of the HDR frame to equal the peak absolute luminance of the display [50,51]. How- ever, such a solution would result in flickering and reduced overall luminance coherence of the HDR video due to the frame-based mapping. as a result be underexposed as it does not correspond to the physical absolute luminance values produced by the display. Grading is therefore required to convert the scene-referred data to output-referred data in order for HDR pixel values to represent the absolute luminance values emitted from the tar- get display. This is also needed as the most HDR metrics are sensitive to the absolute luminance values making grading essential for a correct evaluation [50]. The simplest way of grading is to multiply each HDR frame with a constant that
Table 2 Characteristics of HDR test sequences translates the peak luminance of the HDR frame to equal the peak absolute luminance of the display [50,51]. How- ever, such a solution would result in flickering and reduced overall luminance coherence of the HDR video due to the frame-based mapping. as a result be underexposed as it does not correspond to the physical absolute luminance values produced by the display. Grading is therefore required to convert the scene-referred data to output-referred data in order for HDR pixel values to represent the absolute luminance values emitted from the tar- get display. This is also needed as the most HDR metrics are sensitive to the absolute luminance values making grading essential for a correct evaluation [50]. The simplest way of grading is to multiply each HDR frame with a constant that
Algorithm 1: Pseudocode for grading HDR content.
Algorithm 1: Pseudocode for grading HDR content.
Table3 Rate points for HDR test sequences
Table3 Rate points for HDR test sequences
Fig.6 Tone mapped frames of HDR video test sequences Tunnel (first row), Students (second row) and Window (third row)
Fig.6 Tone mapped frames of HDR video test sequences Tunnel (first row), Students (second row) and Window (third row)
Table 4 Display Use Case 1; average improvement in HDR Quality of HDR video coding methods in comparison with the HLG HDR-TV methoc
Table 4 Display Use Case 1; average improvement in HDR Quality of HDR video coding methods in comparison with the HLG HDR-TV methoc
To illustrate the impact of the HDR Class 1 and HDR Class 2 content, the quality vs. rate performance of six sequences (graded for the display use case 1) are presented in Fig. 8. Here, the sequences Sun, Market3 and BBC02 belong to the HDR class 2, and the other three test sequences, Hallway, Exhibition and Balloon, lie in the HDR class 1 as seen in Table 2.
To illustrate the impact of the HDR Class 1 and HDR Class 2 content, the quality vs. rate performance of six sequences (graded for the display use case 1) are presented in Fig. 8. Here, the sequences Sun, Market3 and BBC02 belong to the HDR class 2, and the other three test sequences, Hallway, Exhibition and Balloon, lie in the HDR class 1 as seen in Table 2.
Fig.9 Average improvement in the HDR quality with respect to the HLG HDR-TV coding method
Fig.9 Average improvement in the HDR quality with respect to the HLG HDR-TV coding method
Related topics:
Computer ScienceArtificial IntelligenceComputer VisionVideo Quality
580 California St., Suite 400
San Francisco, CA, 94104
© 2025 Academia. All rights reserved