Constant Quality Video Encoding

Constant quality encoding targets an encoder-defined quality level while allowing bitrate and output size to vary.

When preparing a library viewing copy, one possible objective is to preserve an acceptable appearance and let the resulting size follow the source. Constant quality encoding starts with that kind of quality policy instead of requiring the video to spend exactly a chosen amount of data.

The policy belongs to the encoder. It guides reconstruction choices; it is not a measurement stamped onto every finished scene.

A quality target makes bitrate the result

The encoder analyzes picture structure and predicts how difficult regions will be to represent. It chooses prediction, partition and quantization decisions that serve its quality objective.

A largely static region may need only a short correction. Changing fine texture can need many significant coefficients. The same quality policy can therefore produce a small description of one region and a larger description of another.

After coding those choices, the number of produced bits is the result. Across a film, varying source difficulty makes bitrate fluctuate and determines the final size.

This reverses an average-budget control priority:

Objective Held as the main target Allowed to respond to content
Average bitrate Overall data budget Reconstruction compromise and local allocation
Quality-led encoding Encoder-defined quality policy Overall size and local bitrate
Fixed quantizer mode Selected quantizer policy Bits and scene-dependent visible error

The last row is related, but not identical to the middle one. Holding one quantizer does not ensure equal perceptual quality across different scenes.

Preserve the parts that matter to later pictures

A reference region can influence more than its own displayed picture. If later predictions reuse it repeatedly, improving that reconstruction can benefit many dependent pictures.

An encoder can incorporate that importance into its quantization decisions. Regional adaptation can also distribute precision differently between smooth areas, detail, and motion according to the chosen model.

These operations explain why the same quality setting does not require a single unchanging QP. CRF, hardware target-quality controls, and CQP have different mechanisms and scales.

The model need not be an external score such as VMAF. Its internal coding estimates and perceptual priorities guide the encode; a later quality measurement evaluates the resulting pictures using another defined method.

A delivery cap adds a competing requirement

Suppose the quality-led choices for a difficult section would require more coded data than its permitted rate and buffer deadline. The encoder cannot both transmit that full representation and satisfy the restrictive deadline.

It must change the picture description: coarser quantization, another prediction/partition choice, or other permitted rate-control behavior. The desired quality becomes a goal constrained by delivery. VBV provides the buffer model behind that decision.

The source and chosen cap determine when this tradeoff occurs. Without the cap, the encode can use more data for that section; with it, the output has another obligation besides the quality policy.

Evaluate the policy on the actual source

A useful evaluation includes material that exercises the policy differently: static detail, grain, rapid movement and smooth gradients. Compare candidate settings using the same source and encoder configuration, looking at the reconstruction as well as its data demand.

For an Emby viewing copy, this establishes both the picture you want to retain and the stream the playback route must supply. Size is then an observed outcome of those decisions, rather than a quality value selected independently of the source.

See Also

References