Video Compression Glossary & Industry Statistics

Entropy coding is a lossless compression technique that assigns shorter binary codes to more frequently occurring symbols in a data stream. In video compression, the entropy coding stage is the final step of every major codec — it is where the statistical redundancy in the encoded data is removed. Common entropy coding methods used in video include CABAC (Context-Adaptive Binary Arithmetic Coding) in H.264 and H.265, and range coding in AV1.

Entropy coding typically accounts for 15–30% of total bitrate savings in a standard encoding pipeline. It cannot remove redundancy that was not present in the data fed to it — which is why pre-processing techniques like entropy conditioning can yield additional gains by improving the data's statistical structure before the codec's entropy stage runs.

Entropy conditioning is a pre-processing technique, originated by MForja, that restructures the statistical distribution of a video or audio stream before it reaches the encoder. By conditioning the stream's entropy profile, the downstream codec's entropy coding stage achieves higher efficiency — resulting in smaller output files without reducing visual quality or altering the codec's own compression logic.

Unlike standard encoder tuning (which adjusts codec parameters), entropy conditioning operates below the codec layer. The output is fully standard-compliant: decoders, CDNs, and playback devices require no changes. MForja's entropy conditioning achieves 50%+ file size reduction across H.264, H.265, AV1, and other codecs.

Video Distillation is MForja's name for the category of technology that removes signal waste from video files before or during encoding. The term deliberately distinguishes this approach from "compression" — which implies degrading the signal — because Video Distillation removes only data that viewers cannot perceive, preserving all perceptually relevant content.

A video file distilled by MForja is 50%+ smaller, visually equivalent to the original, and fully compatible with all standard decoders and delivery infrastructure. No quality settings are changed, no resolution is reduced, and no codec parameters are altered.

Signal waste refers to the data present in a video or audio file that carries no perceptual value to the viewer or listener — bits that encode information the human visual or auditory system cannot distinguish from their absence. Every encoded video file contains signal waste; standard codecs reduce it but do not eliminate it.

MForja's entropy conditioning process identifies and removes signal waste at the entropy layer before encoding, yielding smaller files with no perceptible quality difference. The concept parallels how JPEG progressive encoding removes high-frequency detail that the eye cannot resolve at typical viewing distances.

Lossless compression reduces file size without discarding any data — the original can be reconstructed exactly from the compressed version. Lossy compression achieves higher compression ratios by permanently discarding data deemed perceptually irrelevant. All major video codecs (H.264, H.265, AV1) use lossy compression.

MForja's entropy conditioning is lossless with respect to perceptual quality — no data the viewer can perceive is removed — while achieving compression ratios comparable to lossy techniques.

A codec-agnostic compression approach works across multiple video codecs without requiring the use of a specific encoder or decoder. The compression logic operates independently of the codec — sitting above it as a pre-processor, or below it at the bitstream level — so the same efficiency gains apply whether the pipeline uses H.264, H.265/HEVC, AV1, VP9, or future codecs.

The key benefit is infrastructure stability: adopting a codec-agnostic solution requires no decoder changes, no CDN reconfiguration, and no migration risk. Organizations can improve compression efficiency without committing to a codec transition.

A pre-processing shim is a software layer inserted into a media pipeline upstream of an encoder. It processes the video or audio stream before encoding begins, applying transformations that improve encoding efficiency, reduce noise, or condition the signal for better downstream compression. The shim is transparent to the encoder itself — the encoder processes the shim's output as it would any other input.

Pre-processing shims require no changes to the encoder, decoder, CDN, or playback devices. They are among the least disruptive ways to improve pipeline efficiency because they can be inserted or removed without altering any other component of the stack.

Rate control is the mechanism by which a video encoder allocates bits across a video stream to meet a target bitrate or quality level. Common rate control modes include CBR (Constant Bitrate), VBR (Variable Bitrate), CRF (Constant Rate Factor), and CQP (Constant Quantization Parameter). Rate control determines how much quality is traded for file size at each frame.

When a pre-processing step like entropy conditioning reduces the informational complexity of the stream, rate control can achieve the same quality targets at lower bitrates — meaning smaller files without adjusting any rate control settings.

Perceptual quality refers to how a compressed video looks to a human viewer, as opposed to how closely it matches the original at a mathematical level. Two videos can have very different PSNR (Peak Signal-to-Noise Ratio) scores but appear identical to human observers — and vice versa. Perceptual quality is the standard by which consumer and enterprise video quality is evaluated.

The key perceptual quality metrics used in industry are VMAF (developed by Netflix), SSIM (Structural Similarity Index), and PSNR. MForja's entropy conditioning is designed to preserve perceptual quality — VMAF and SSIM scores are maintained at levels indistinguishable from the source.

VMAF is a perceptual video quality metric developed by Netflix and now widely adopted across the streaming industry. It combines multiple image quality metrics (including VIF, DLM, and motion) into a single score that correlates strongly with how humans perceive quality. VMAF scores range from 0–100; scores above 93 are generally considered indistinguishable from source quality by human viewers.

VMAF has become the industry standard for validating that compression techniques preserve quality, replacing PSNR-only assessments which poorly correlate with human perception. It is particularly useful for evaluating pre-processing techniques, where traditional metrics may not capture the full picture.

H.265 (also called HEVC — High Efficiency Video Coding) is the successor to H.264, standardized in 2013. It achieves approximately 40–50% better compression than H.264 at the same quality level by using larger and more flexible coding unit structures (up to 64×64 pixels vs. H.264's 16×16), improved motion compensation, and more sophisticated entropy coding via CABAC.

Deploying H.265 requires updates to both encoders and decoders. Many older devices, browsers, and CDN configurations do not support H.265 natively, making migration a significant infrastructure project. H.265 licensing fees also remain a barrier for some deployments — a factor that accelerated adoption of the royalty-free AV1 codec.

AV1 is a royalty-free open video codec developed by the Alliance for Open Media (AOMedia), released in 2018. It achieves approximately 30% better compression than H.265 at equivalent quality, with no licensing fees. AV1 is supported natively in Chrome, Firefox, and newer versions of Safari, and is the codec of choice for YouTube and Netflix for compatible devices.

AV1's main limitation is encoding speed — it requires significantly more compute than H.264 or H.265 to encode, making it challenging for live or real-time workflows without dedicated hardware encoders. Decoder support is still growing, particularly on older mobile devices.

A Group of Pictures (GOP) is the sequence of video frames between two I-frames (independently encoded keyframes). The GOP structure determines how frequently full keyframes appear in the stream. A shorter GOP (e.g., every 30 frames at 30fps = 1 second) improves seek accuracy and resilience but increases file size because more I-frames must be encoded. A longer GOP improves compression but reduces random access performance.

For streaming delivery, GOP length is a key tuning parameter. HLS and DASH typically require I-frames at segment boundaries (every 2–6 seconds), which effectively sets a maximum GOP length for adaptive bitrate delivery.

Bitrate is the number of bits used to represent one second of video or audio, measured in kilobits per second (kbps) or megabits per second (Mbps). Higher bitrate generally means higher quality and larger file sizes. For a given codec and content type, there is a minimum bitrate below which quality degrades noticeably — and a ceiling above which additional bits provide no perceptible quality improvement.

Bitrate directly determines both file size (bitrate × duration = file size) and CDN bandwidth consumption. Reducing bitrate without reducing quality — which MForja's entropy conditioning enables — is the most direct way to reduce both storage and delivery costs simultaneously.

CDN egress (or "data transfer out") is the cost charged by a Content Delivery Network for transmitting data from its edge servers to end users. It is typically the largest variable cost for video streaming platforms, charged per gigabyte delivered. Enterprise CDN egress pricing typically ranges from $0.01–$0.12 per GB depending on volume, region, and provider.

Because CDN egress is priced per gigabyte, file size reductions translate directly and proportionally into cost reductions. A 50% reduction in encoded file size produces approximately a 50% reduction in CDN egress spend for equivalent viewer hours.

Object storage is a cloud storage architecture that stores data as discrete objects (files) in a flat namespace, accessed via API. The major object storage services — AWS S3, Google Cloud Storage, Azure Blob Storage — are the primary storage layer for video libraries, surveillance archives, and media assets at scale. Pricing is charged per gigabyte stored per month.

At scale, object storage costs for video are substantial. AWS S3 Standard pricing is approximately $0.023/GB/month. A 2-petabyte video library costs approximately $46,000/month — $552,000/year — just in storage. Reducing file sizes by 50% cuts that to $276,000/year.

Transcoding is the process of converting a video file from one codec, resolution, bitrate, or format to another. It involves decoding the source file and re-encoding the output — a computationally expensive process. Large-scale transcoding pipelines (for OTT platforms, UGC platforms, surveillance systems) consume significant GPU and CPU resources, and GPU compute costs are a major operational expense.

Smaller input files reduce transcoding time and cost proportionally. When MForja's entropy conditioning reduces file sizes by 50%, the transcoding pipeline processes half the data for the same output — reducing GPU hours, cloud compute costs, and carbon footprint by a comparable margin.

Adaptive Bitrate (ABR) streaming is a delivery technique that encodes the same video at multiple quality levels (bitrate ladders), allowing players to dynamically switch between quality tiers based on available bandwidth and device capability. ABR is the standard method for delivering video over the internet — it is the technology behind Netflix, YouTube, Disney+, and virtually every major streaming platform.

An ABR bitrate ladder for a single piece of content might include 5–10 renditions ranging from 240p/400kbps to 4K/15Mbps. Each rendition must be encoded, stored, and delivered independently — multiplying the storage and egress cost impact of any file size optimization.

HLS (HTTP Live Streaming) is Apple's adaptive streaming protocol, originally developed for iOS and now universally supported. DASH (Dynamic Adaptive Streaming over HTTP) is the international standard equivalent (ISO/IEC 23009). Both segment video into small chunks (typically 2–6 seconds each) and provide manifest files that players use to select and download the appropriate quality tier.

HLS is the dominant format for mobile and OTT delivery; DASH is preferred for DRM-protected content on Android and connected TVs. Most enterprise and consumer streaming platforms support both. Segment size directly affects latency, seek performance, and storage overhead — smaller encoded segments reduce both.

A container format (or wrapper) is the file format that holds encoded video, audio, subtitles, and metadata together. Common containers include MP4 (MPEG-4 Part 14), MOV (QuickTime), MKV (Matroska), and TS (MPEG Transport Stream). The container is distinct from the codec — an MP4 file can contain H.264, H.265, or AV1 video; the container simply packages whatever codec is inside.

Container format is relevant to CDN delivery, device compatibility, and DRM support, but has minimal impact on compression efficiency. A file's codec and encoding settings — not its container — determine its size and quality characteristics.