ISO/IEC 10918-1 : 1993(E)
Before error recovery procedures can be invoked, the error condition must first be detected. Errors during decoding can
show up in two places:
The decoder fails to find the expected marker at the point where it is expecting resynchronization.
Physically impossible data are decoded. For example, decoding a magnitude beyond the range of values
allowed by the model is quite likely when the compressed data are corrupted by errors. For arithmetic
decoders this error condition is extremely important to detect, as otherwise the decoder may reach a
condition where it uses the compressed data very slowly.
NOTE Some errors will not cause the decoder to lose synchronization. In addition, recovery is not
possible for all errors; for example, errors in the headers are likely to be catastrophic. The two error
conditions listed above, however, almost always cause the decoder to lose synchronization in a way which
In regaining synchronization, the decoder can make use of the modulo 8 coding restart interval number in the low order
bits of the RST
marker. By comparing the expected restart interval number to the value in the next RST
marker in the
compressed image data, the decoder can usually recover synchronization. It then fills in missing lines in the output data by
replication or some other suitable procedure, and continues decoding. Of course, the reconstructed image will usually be
highly corrupted for at least a part of the restart interval where the error occurred.
Sequential DCT decoding process with Huffman coding and 12-bit precision
This process is identical to the sequential DCT process defined for 8-bit sample precision and extended to four Huffman
tables, as documented in F.2.3, but with the following changes.
Structure of DC Huffman decode table
The general structure of the DC Huffman decode table is extended as described in F.1.5.1.
Structure of AC Huffman decode table
The general structure of the AC Huffman decode table is extended as described in F.1.5.2.
Sequential DCT decoding process with arithmetic coding and 12-bit precision
The process is identical to the sequential DCT process for 8-bit precision except for changes in the precision of the IDCT
The structure of the decoding procedure in F.2.4 is already defined for a 12-bit input precision.
CCITT Rec. T.81 (1992 E)