When seeking within a video file, the playhead jumps forward or back rather than playing from the specified point.
To enable accurate seeking and scrubbing within a video you must deliver the video via RTMP streaming with Flash Media Server or a Flash Video Streaming Service. With progressive delivery, you must embed a navigation cue points at specified timecodes to allow seeking to that point. Seeking is also possible with HTTP Dynamic Streaming, but is still not frame-accurate.
Accurate seeking requires the existence of a keyframe at the seek point. A keyframe is a video frame that contains 100% of the pixel information for that frame image.
Flash Media Server can generate keyframes on the fly, so streaming via RTMP allows seeking to any chosen frame.
With progressive delivery, seeking is restricted to downloaded bits, and also to encoded keyframes. This will cause the playhead to jump to the nearest encoded keyframe rather than start playback at the chosen point. If you want viewers to be able to seek to a specific point in progressive video -- at a scene change, for example -- you can embed a navigation cue point at that timecode, forcing the encoder to place a whole keyframe at that point.
HTTP Dynamic Streaming can enable more accurate seeking than progressive delivery, but is also restricted to encoded keyframes. This delivery method breaks the video up into fragments, with each fragment being a Group of Pictures (GOP) starting with a keyframe. Seeking occurs to the closest GOP. The accuracy of seeking is therefore dependent on fragment size. Requested fragments can begin playing back right away, so the video does not need to be downloaded to the seek point as required with progressive delivery.
For more information, visit http://www.adobe.com/devnet/flash/learning_guide/video/part03.html