ISO/IEC 10918-1 : 1993(E)
In the hierarchical mode the define-hierarchical-progression (DHP) marker segment shall be placed in the compressed
image data before the first start-of-frame. The DHP segment is used to signal the size of the image components of the
completed image. The syntax of the DHP segment is specified in Annex B.
The first frame for each component or group of components in a hierarchical process shall be encoded by a
non-differential frame. Differential frames shall then be used to encode the two's complement differences between source
input components (possibly downsampled) and the reference components (possibly upsampled). The reference
components are reconstructed components created by previous frames in the hierarchical process. For either differential or
non-differential frames, reconstructions of the components shall be generated if needed as reference components for a
subsequent frame in the hierarchical process.
Resolution changes may occur between hierarchical frames in a hierarchical process. These changes occur if
downsampling filters are used to reduce the spatial resolution of some or all of the components of the source image. When
the resolution of a reference component does not match the resolution of the component input to a differential frame, an
upsampling filter shall be used to increase the spatial resolution of the reference component. The EXP marker segment
shall be added to the compressed image data before the start-of-frame whenever upsampling of a reference component is
required. No more than one EXP marker segment shall precede a given frame.
Any of the marker segments allowed before a start-of-frame for the encoding process selected may be used before either
non-differential or differential frames.
For 16-bit input precision (lossless encoder), the differential components which are input to a differential frame are
calculated modulo 2
. The reconstructed components calculated from the reconstructed differential components are also
calculated modulo 2
If a hierarchical encoding process uses a DCT encoding process for the first frame, all frames in the hierarchical process
except for the final frame for each component shall use the DCT encoding processes defined in either Annex F or Annex
G, or the modified DCT encoding processes defined in this annex. The final frame may use a modified lossless process
defined in this annex.
If a hierarchical encoding process uses a lossless encoding process for the first frame, all frames in the hierarchical process
shall use a lossless encoding process defined in Annex H, or a modified lossless process defined in this annex.
The downsampled components are generated using a downsampling filter that is not specified in this Specification. This
filter should, however, be consistent with the upsampling filter. An example of a downsampling filter is provided in K.5.
The upsampling filter increases the spatial resolution by a factor of two horizontally, vertically, or both. Bi-linear
interpolation is used for the upsampling filter, as illustrated in Figure J.2.
Figure J.2 Diagram of sample positions for upsampling rules
Figure J.2 [D109] = 4 cm = 156 %
CCITT Rec. T.81 (1992 E)