Kenneth Vermeirsch

Efficient spatial resolution reduction transcoding for H.264/AVC

In this paper, we present a spatial resolution reduction transcoding architecture for H.264/AVC, which extends open-loop transcoding with a low-complexity compensation technique in the reduced-resolution domain. The proposed architecture removes visual artifacts from the transcoded sequence, while keeping complexity significantly lower than more traditional cascaded decoder-encoder architectures. The refinement step of the proposed architecture can be used to further improve rate-distortion performance, at the cost of additional complexity. In this way, a dynamic-complexity transcoder is rendered possible.

Garbageless Reversible Implementation of Integer Linear Transformations

Discrete linear transformations are important tools in information processing. Many such transforms are injective and therefore prime candidates for a physically reversible implementation into hardware. We present here reversible digital implementations of different integer transformations on four inputs. The resulting reversible circuit is able to perform both the forward transform and the inverse transform. Which of the two computations that actually is performed, simply depends on the orientation of the circuit when it is inserted in a computer board (if one takes care to provide the encapsulation of symmetrical power supplies). Our analysis indicates that the detailed structure of such a reversible design strongly depends on the prime factors of the determinant of the transform: a determinant equal to a power of 2 leads to an efficient garbage-free design.

Dyadic spatial resolution reduction transcoding for H.264/AVC

In this paper, we examine spatial resolution downscaling transcoding for H.264/AVC video coding. A number of advanced coding tools limit the applicability of techniques, which were developed for previous video coding standards. We present a spatial resolution reduction transcoding architecture for H.264/AVC, which extends open-loop transcoding with a low-complexity compensation technique in the reduced-resolution domain. The proposed architecture tackles the problems in H.264/AVC and avoids visual artifacts in the transcoded sequence, while keeping complexity significantly lower than more traditional cascaded decoder–encoder architectures. The refinement step of the proposed architecture can be used to further improve rate-distortion performance, at the cost of additional complexity. In this way, a dynamic-complexity transcoder is rendered possible. We present a thorough investigation of the problems related to motion and residual data mapping, leading to a transcoding solution resulting in fully compliant reduced-size H.264/AVC bitstreams.

Increased flexibility in inter picture partitioning

To attain efficient coding of sequences with complex motion activity, modern video coding standards allow variable block sizes to be employed in temporal prediction. A block with complex motion can be partitioned into two equal-sized halves or into four quadrants. In this paper we study the impact of allowing blocks to be partitioned in two unequal subpartitions. Additionally, we allow block partitioning along diagonal edges. These provisions allows encoders to better adapt to the local characteristics of the motion activity in a video sequence. We verified experimentally that the presence of partitioning edges that do not coincide with transform boundaries does not negatively impact the decorrelating strength of the residual transform, so the proposed extended partitioning strategies can be applied regardless of the details of the residual coder. Implementing the proposed extended partitioning modes in an H.264/AVC coder at the macroblock and submacroblock level, we observe that coding efficiency gains are greatest in low-resolution sequences, where moving features in the video sequence tend to be more spatially localized. For CIF and QCIF sequences we achieve a bit rate reduction of about 3-6% over a wide fidelity range.

Efficient adaptive-shape partitioning of video

While many recent international video coding standards, especially H.264/MPEG-4xa0AVC, leverage block size adaptivity in motion estimation, the rate-distortion boundary can be pushed further by allowing even more freedom in the partitioning process of inter pictures. Adaptive-shape partitioning, which allows blocks to be partitioned along a straight line that runs through the block at a freely chosen angle and position, complements the regular subblock partitioning, allowing the encoder to better adapt to the local characteristics of the motion activity in a video sequence. However, the technique demands excessive encoder resources to exhaust the large search space. This paper is the result of an investigation into the relative rate-distortion importance of the various adaptive-shape modes, both in terms of the angle of the partition boundary and of its location within a block. We find that a significant reduction of the search space with a factor of up to 40 can be accomplished, while retaining 50 to 90% of the compression gain obtained in the state of the art. This allows encoders to operate at much lower complexity levels and also reduces the signaling overhead associated with adaptive-shape partitioning. Based on our observations, we formulate a number of approaches to trade off compression performance against encoder complexity. Furthermore we discuss the use of various schemes of overlapping motion estimation along the partition boundary, an aspect which is currently left unaddressed in the literature on adaptive-shape partitioning. We introduce the use of shape-adaptive transforms for the motion compensated signal, to avoid the condition that arises with adaptive-shape partitioning where a partition boundary lies inside a transform block. The result is a reduction in ringing artifacts while maintaining objective quality.

Evaluation of transform performance when using shape-adaptive partitioning in video coding

When combining non-rectangular (shape-adaptive) partitioning of inter pictures with a rectangular block transform, some of the transforms will be applied to a residual signal which originates from two different predictions. This is a condition that is never encountered in any of the existing video coding standards, and the impact of this on the performance of the transform coder has not been investigated. In this paper we investigate the effect of these mixed-signal blocks on the coding gain for various transforms, and we compare these against the optimal KLT gain. We find that, despite the transformed residual block being composed of two parts that were predicted from different areas of the reference picture, correlation within mixed blocks is very similar to that of normal blocks. The DCT is only marginally suboptimal w.r.t. KLT. KLT has practical issues that will reduce its coding gain or increase the signaling overhead: transform bases need to be quantized and transmitted, or a number of fixed bases needs to be chosen offline. Therefore we recommend DCT be used for all types of blocks.

Leveraging the quantization offset for improved requantization transcoding of H.264/AVC video

Requantization transcoding is a method for reducing the bit rate of compressed video bitstreams. Most research on requantization is concerned with the architectural design, the selection of a suitable quantizer, or the reduction of requantization errors. In this paper, we propose to incorporate a new dimension in requantization transcoding: the quantization offset. We compare two requantization methods: increasing the quantization step size and decreasing the quantization offset. Furthermore, we propose a novel heuristic for requantization transcoding based on a theoretical rate-distortion analysis. The experimental results for H.264/AVC video show that requantization is improved in rate-distortion sense with gains up to 1 dB for open-loop requantization of B pictures compared to requantization with fixed quantization offset.

Transcoding of H.264/AVC to SVC with motion data refinement

In this paper, we present motion-refined transcoding of H.264/AVC streams to SVC in the transform domain. By accurately taking into account both rate and distortion in the different layers on the one hand, and the SVC inter-layer motion prediction mechanisms on the other hand, the proposed transcoding architecture is able to improve rate-distortion performance over existing approaches. We propose a multilayer control mechanism that trades off performance between the different layers, resulting in 0.5 dB gains in the output SVC base layer.

Low Complexity Multiple Description Coding for H.264/AVC

Todays wireless networks suffer from high error rates in real-life conditions. Moreover errors tend to occur in large bursts. Multiple Description Coding (MDC) of digital video is a promising technique to overcome these problems in wireless environments by taking advantage of frequency or channel diversity. This papers proposes two low complexity MDC techniques for H.264/AVC at the Network Abstraction Layer level, aiming at conversational (video teleconferencing) and sports Video on Demand (VoD) application scenarios, respectively. For conversational applications, an MDC scheme based on H.264/AVC data partitioning is used, offering low additional delay and good performance for video with static backgrounds. For the sports VoD application, a scheme based on redundant slices is proposed. The latter scheme performs better for high-motion sequences than the former, but also introduces more delay.

Region-adaptive probability model selection for the arithmetic coding of video texture

In video coding systems using adaptive arithmetic coding to compress texture information, the employed symbol probability models need to be retrained every time the coding process moves into an area with different texture. To avoid this inefficiency, we propose to replace the probability models used in the original coder with multiple switchable sets of probability models. We determine the model set to use in each spatial region in an optimal manner, taking into account the additional signaling overhead. Experimental results show that this approach, when applied to H.264/AVCs context-based adaptive binary arithmetic coder (CABAC), yields significant bit-rate savings, which are comparable to or higher than those obtained using alternative improvements to CABAC previously proposed in the literature.