Adaptive Re-Quantization For High Dynamic Range Video Compression |
contact: M. Le Pendu, C. Guillemot, D. Thoreau
High Dynamic Range (HDR) images contain more intensity levels than traditional image formats. Instead of 8 or 10 bit integers, floating point values are generally used to represent the pixel data. To extend the use of existing video codecs such as HEVC to HDR floating point video sequences, we propose a method that converts the floating point data and reduces the bit depth of input images with minimal loss. Several variants of the method are proposed. They are adapted to different quality requirements. In particular, near lossless compression is addressed.
The coding scheme is illustrated in Fig. 1. The input 16 bits half float RGB data is first converted to integer without loss, by taking the bit pattern of the floating point numbers (ignoring the sign bit) and interpreting them as integers. We obtain 15 bit integers that approximately correspond to a logarithmic encoding of the original values. After applying a conversion to YUV colorspace, the bitdepth is reduced to fit the encoder requirements. This operation is detailed in Fig.2. For a target bitdepth n, it consists in a linear scaling that maps the minimum and maximum values of the data to 0 and 2^{n}-1 respectively. The scaling is followed by a rounding operation to obtain n bit integers. Three variants are proposed : the algorithm can be performed either by blocks, by frame or by group of picture (GOP). Finally, the resulting image is encoded with HEVC. The minimum and maximum values are necessary to reconstruct the final HDR image and must also be encoded.
Fig. 1 - General coding scheme used for floating point HDR image encoding.
Fig. 2 - Adaptive Quantization based on minimum and maximum values.
The three variants of the method are adapted to different usages :
For the experiments, we have considered the sequence Tunnel video clip ( 1080p format ). The group of pictures (GOP) size is set to eight for the tests using inter predictions. The encoding is performed using the HEVC range extension version using up to 14 bit target bitdepth and YUV444 chroma format.
A wide range of QP from 0 to 50 was used to generate the rate distortion curves in figures 4 and 5 and negative QP values were also used in figure 6 to show the performance of the block-wise method in conditions of near lossless compression.
Fig. 3 - One frame of the tested Tunnel video clip sequence. |
Fig. 4 - Rate distortion results for GOP-wise method. |
Fig. 5 - Rate distortion results for Frame-wise method. |
Fig. 6 - Comparison between 14 bit frame-wise, 12 bit block-wise and adaptive LogLuv methods. |