PxPixel
Harmonic's MV50 MPEG-2 encoder - TvTechnology

Harmonic's MV50 MPEG-2 encoder

Integrating noise reduction and video preprocessing functions with the hardware that performs video compression offers key performance advantages as well as the obvious savings of rack space, power and cooling support. Encoding noise is a waste of bandwidth. Removing noise enables improved video delivery at lower bit rates.
Author:
Publish date:

Integrating noise reduction and video preprocessing functions with the hardware that performs video compression offers key performance advantages as well as the obvious savings of rack space, power and cooling support. Encoding noise is a waste of bandwidth. Removing noise enables improved video delivery at lower bit rates.

Noise shares the same space as valuable picture detail, and levels of noise can vary dramatically. This is especially true with archive material and news feeds. A solution is to deploy multiple preprocessing filters, each with a specific role, and then combine them with an automatic control system to apply the optimal filter at the right strength in the right places. The final objective of the video preprocessing and compression operation is to maximize the subjective quality of the decoded video signal for a given encoded bit rate.

Meeting the challenge

Harmonic’s MV50 MPEG-2 encoder now supports its third evolution of noise reduction (NR) technology. A central feature is “LookAhead” architecture, a multi-processor combination designed to aid both NR and compression by providing analysis of the incoming content before it meets the master compression engine. A dedicated MPEG-2 video signal processor is used to extract input signal statistics that are used to assist the decision processes by the compression engine and preprocessing filters.

The MV50 employs a suite of preprocessing filters that perform spatial, temporal and spatio-temporal processing of the input video signal.

Motion Compensated Temporal Filtering (MCTF) is a technique used to remove/attenuate Gaussian noise from video content. This form of filtering is much more powerful than temporal-only filtering, in which pixels are filtered with other pixels at the same spatial location. With MCTF, filtering is applied along “motion trajectories.” If an object within a sequence moves from frame to frame, block-based motion estimation (ME) is used to track the direction and magnitude of the motion. Filtering is then applied using pixels that retain the same position relative to an object moving within the frame sampling structure.

MCTF all but eliminates ghosting and trailing normally associated with temporal-only filters. Random components are heavily suppressed, yet picture details are preserved. It also limits artifacts to within the threshold of visibility.

Simple spatial filtering applies intra-frame, two-dimensional low-pass filtering to suppress high-frequency noise. The downside of this technique is less picture detail. Viewers generally have limited tolerance for reduction in detail and edge definition; therefore, basic spatial filtering should be applied sparingly. Harmonic has developed a new spatial technique that is naturally edge preserving but allows significant filtering to remove noise with Gaussian characteristics. This filter’s operation is coupled to the MCTF when motion within a sequence is not well-approximated by the translational block motion model, or when it has been determined that the input sequence contains a strong Gaussian-type noise component.

Impulse noise occurs for a single frame and is commonly seen as dots, scratches or film damage. This type of noise does not waste bits like random noise, but it has a significant effect on perceived picture quality and reduces the effectiveness of other noise reduction and compression mechanisms.

The MV50 encoder devotes two thousand million pixel comparisons per second to ascertaining whether a pixel, or group of pixels, are single-frame rogues. Once the impulses have been identified, it is relatively easy to substitute them using values based on spatio-temporal filtering operations.

The Edge Adaptive Texture (EAT) filter is a spatio-temporal filter used to drive down bit rates, extending the encoder’s performance. It is well-known that progressive scan video material is more efficiently encoded than interlace video. Interlaced video can be made more efficient by performing an interlace-to-progressive spatio-temporal format conversion. Unfortunately, this operation introduces motion artifacts when viewed on an interlaced television.

The EAT filter has been designed to perform an adaptive de-interlacing function on the input interlaced video signal. Filtering is performed on regions identified as texture. However, edge-defining objects, or those containing regions of texture, are left in their native interlaced format. Filtering the textured regions in this way can provide the benefits of improved coding efficiency – much like those realized by de-interlacing – without the drawbacks. When this preprocessing technique is applied to sequences containing high spatial detail and motion, significant improvements in low bit rate encoding efficiency are obtained.

The combination of advanced compression technology and powerful and innovative preprocessing techniques is enabling enhanced delivery of digital content at ever-lower bit rates.

Dr. Andrew W. Johnson is staff development engineer and Neil Brydon is senior group marketing manager for the Convergent Systems Division of Harmonic Inc.