Analysis of one-dimensional transforms in coding motion compensation prediction residuals for video applications

Abstract

In video coding, motion compensation prediction provides significant increases in overall compression efficiency. The prediction residuals are typically treated as images and compressed by applying two-dimensional transforms such as the two-dimensional discrete cosine transform (2D-DCT). Previous work has found that the use of directionadaptive one-dimensional discrete cosine transforms (1D-DCTs) in coding motion compensation residuals can provide significant additional bitrate savings. However, this requires optimization over all of the available transforms to minimize the overall bitrate, which can be expensive in terms of time and computation.In this thesis, we examine the use of only the horizontal and vertical 1D-DCTs in addition to the 2D-DCT for coding motion compensation residuals. By reducing the number of available transforms, the amount of required computation decreases significantly, with a potential cost in performance. We perform experiments using a modified H.264/AVC codec to compare the performance of using different sets of available transforms. The results indicate that for typical applications of video coding, most of the performance benefit from using directional 1D-DCTs can be retained by keeping only the horizontal and vertical 1D-DCTs.

Algorithms for transform selection in multiple-transform video compression

“…Table 4.2 shows the relative coefficients saving with respect to using only 2D-DCT. These results are consistent with the Bjontegaard-Delta bitrate results reported in [5]. The coefficients saving, as well as the bitrate saving, increases when more transforms are used.…”

“…Qualitatively, this result is consistent with the rate-distortion performance reported in [5], where these transforms are incorporated into a H.264 video coding system. The preserved energy is related with the distortion present in a compressed video sequence, and the number of coefficients used roughly indicates the number of bits in coding a residual image.…”

“…The work in [2] reports that with additional 1D-DCTs of 16 different directions, the rate-distortion performance of a video coding system significantly increases. The work in [5] analyzes the performance of these 1D-DCTs, and reports that much of the performance gain by using the 1D-DCTs is due to the use of only horizontal and vertical 1D-DCTs. In this section, we evaluate the energy compaction of three transforms sets.…”

Transforms for Intra Prediction Residuals Based on Prediction Inaccuracy Modeling

“…In the work reported in [14]- [16], it is observed that many one-dimensional anisotropic structures arise in motioncompensated residuals. By modeling the motion-compensated residuals with one-dimensional first-order Markov models along a certain direction, a set of one-dimensional transforms of many directions is derived for encoding motioncompensated residuals.…”

Statistical analysis and directional coding of layer-based HDR image coding residue

“…Although some recent studies [10], [11] design function based global mapping curves for the 8-bit conversion and uses typical codecs for residue coding, they do not consider the local characteristics of residue blocks for encoding. However, in well-studied motion compensated LDR video coding [12], [14], it is verified that the residue blocks exhibit local anisotropic characteristics, which are more efficiently compressed with directional transforms than the traditional 2-D DCT.…”