In MPEG compression, macroblocks with different information content, the Group of Pictures (GOP), are formed based on the similarity of successive video frames. Within these groups of pictures, the pictures are compressed one after the other, starting from the data-intensive initial picture of the group, the intra-frame(I-frame). This frame forms the information basis for the subsequent frames and defines the reference within the video sequence. The I-frame is compressed in a similar way to JPEG.
The following frames are the bidirectional frames (B-frames), because they depend equally on the preceding frames as well as the following ones and use them as reference. They are particularly highly compressed frames because only the differences between them and the neighboring frames serve as information. A B-frame group is followed by a predicted frame (P-frame), which is based on the prediction from the intra-frame. The P-frame is oriented to the reference frame, the I-frame, forms the reference points for the I-frames and is particularly important for decompression.
The length of the GOP frame depends on the video codec setting and ranges from 0.5 seconds for MPEG-2 to over 10 seconds for MPEG-4. In addition to video coding, MPEG compression also includes multi-channel audio coding, which is implemented in MPEG-1 audio and MPEG-2.