Contents Page

7.9 Temporal scalability

The reference frames for prediction are selected by reference_select_code as described in Tables 7-28 and 7-29. In P pictures, the forward reference picture can be one of the following three: most recent enhancement picture, most recent lower layer frame, or next lower layer frame in display order. Note that in the latter case, the reference frame in lower layer used for prediction is backward in time.

In B-pictures, the forward reference can be one of the following two: most recent the enhancement pictures or most recent (or temporally coincident) lower layer frame whereas the backward reference can be one of the following two: most recent lower layer picture including temporally coincident picture in display order or next lower layer frame in display order. Note that in this case, the backward reference frame in lower layer used for prediction is forward in time.

Backward prediction cannot be made from a picture in the enhancement layer. This avoids the need for frame reordering in the enhancement layer. Motion compensation process forms predictions using lower layer decoded pictures and/or previous temporal prediction from the enhancement layer.

The enhancement layer can contain I-pictures, P-pictures or B-pictures, but B-pictures in enhancement layer behave more like P-pictures in the sense that a decoded B-picture can be used to predict the following P-pictures or B-pictures in the enhancement layer.

When the most recent frame in the lower layer is used as the reference, this includes the frame that is temporally coincident with the frame or the first field (in case of field pictures) in the enhancement layer. The prediction references used for P-picture and B-pictures are shown in Table 7-28 and Table 7-29 respectively.

The lower and enhancement layers shall use the restricted slice structure.

Table 7-28 Prediction references selection in P-pictures

reference_select_code	forward prediction reference
00	Most recent decoded enhancement picture(s)
01	Most recent lower layer frame in display order
11	forbidden

Table 7-29 Prediction references selection in B-pictures

reference_select_ code	forward prediction reference	backward prediction reference
00	forbidden	forbidden
01	Most recent decoded enhancement picture(s)	Most recent lower layer picture in display order
11	Most recent lower layer picture in display order	Next lower layer picture in display order

Figure 7-16 shows a simplified diagram of the motion compensation process for the enhancement layer using temporal scalability.



Figure 7-16 Simplified motion compensation process for the enhancement layer using temporal scalability.

I-pictures do not use prediction references; to indicate this, the reference_select_code for I-pictures shall be ‘11’ .

Depending on picture_coding_type, when forward_temporal_reference or backward_temporal_reference do not imply references to be used for prediction, they shall take the value 0.

7.9.1 Higher syntactic structures

The two bitstreams shall have consecutive layer ids, with enhancement layer having layer_id=id_enhance and the lower layer having layer_id=id_enhance-1.

The syntax and semantics of enhancement layers are as defined in Clauses 6.2 and 6.3 respectively.

Semantic restrictions apply to several values in the headers and extensions of the enhancement layer as follows.

The lower layer shall conform to this specification (and not to ISO/IEC 11172-2).

Sequence header

The values in this header can be different from the lower layer except for horizontal_size_value, vertical_size_value and aspect_ratio_information.

This extension shall be identical to the one in the lower layer except for values of profile_and_level_indication, bit_rate_extension, vbv_buffer_size_extension, low_delay, frame_rate_extension_n and frame_rate_extension_d. These can be selected independently. Note that progressive_sequence indicates the scanning format of the enhancement layer frames only rather than of the output frames after multiplexing. The latter is indicated by mux_to_progressive_sequence (see sequence scalable extension).

This extension shall not be present as there is no separate display process for the enhancement layer.

This extension shall be present with scalable_mode = “Temporal scalability”.

When progressive_sequence=0 and mux_to_progressive_sequence=0, top_field_first and picture_mux_factor can be selected.

When progressive_sequence=0 and mux_to_progressive_sequence=1, top_field_first shall contain a complement of the value of top_field_first of the lower layer but picture_mux_factor shall be 1.

When progressive_sequence=1 and mux_to_progressive_sequence=1, top_field_first shall be zero but picture_mux_factor can be selected.

The combination of progressive_sequence=1 and mux_to_progressive_sequence=0 shall not occur.

There is no restriction on GOP header (if present) to be the same as that for lower layer

There is no restriction on picture headers to be the same as in the lower layer.

The values in this extension can be different from the lower layer except for top_field_first, concealment_motion_vectors, and chroma_420_type and progressive_frame. The top_field_first shall be based on progressive_sequence and mux_to_progressive_sequence (see sequence_scalable_extension above) and concealment_motion_vectors shall be 0. Chroma_420_type shall be identical to the lower layer. Progressive_frame shall always have the same value as progressive_sequence.

This extension shall be present for each picture.

This extension may be present in the enhancement layer.

7.9.2 Restrictions on temporal prediction

In case group_of_pictures_header occurs very often in lower_layer, ambiguity can occur due to possibility of nonuniqueness of temporal references (which are reset at each group_of_pictures_header). This ambiguity shall be resolved with help of systems layer timing information.