B-frames are one of three possible frame types as outlined in the MPEG-4 video compression standard. The three frame-types are referred to as I(ntra)-frames, P(redicted)-frames and B(i-directional)-frames.

In such compression formats, such as MPEG-4, only differentiating data between frames is recorded. For example, in a scene whereby a person stands still while lowering his guns, only the motion of him moving his arms will be recorded into those frames (either as motion vector co-ordinates or image data, depending on complexity). The rest of the scene does not have any motion; therefore storing it would be redundant in future frames.

Some knowledge of perceptual video compression will be assumed for the rest of the article.

Intra-frame

• Is a complete image.
• Only contains image data, not motion vector co-ordinates.
• No dependence on reference frames.
• Are generated when differentiating image details prohibit generation of effective P or B frames.
• Used to help decode successive frames.
• I-frames are also known as keyframes because of their importance. Longer intervals between I-frame appearances translates into longer frame reconstruction times, which in turn means longer seek times for local media and longer recovery times for streaming media if the connection goes bad.

Predicted-frame

• Only contains partial data.
• May contain either image data, or motion vector co-ordinates.
• Can only reference frames in a singular direction that all P-frames must follow.
• Are generated when differentiating details can be referenced from the previous relevant frame.
• Used to help rebuild certain frames.
• Compresses better than an I-frame.

Bi-directional frame

• Only contains partial data.
• May contain both image data or motion vector co-ordinates.
• B-frames are never referenced to. Owing to this, higher quantization values can be used for B-frames since there is no risk of quality degradation resulting from rounding-off errors.
• References both the previous and next frame simultaneously.
• Are generated when the algorithm determines bi-directional data to be necessary.
• Used to help rebuild certain frames.
• Most well-suited towards compression among the three types.