Why Baseline JPEG Decoders Fail on Arithmetic Coding

Baseline JPEG decoders fail when encountering an arithmetic-coded JPEG stream primarily because they are engineered to support only Huffman entropy coding, which is the mandatory algorithm defined for the JPEG baseline standard. When presented with an arithmetic-coded stream, these decoders encounter incompatible frame markers, missing Huffman definitions, and an entirely different mathematical method of bitstream representation that they lack the algorithmic logic to interpret.

1. Incompatible Frame Markers (SOF)

The JPEG standard (ITU-T T.81 / ISO/IEC 10918-1) defines different Start of Frame (SOF) markers to identify the encoding process used in the image:

• SOF0 (0xFFC0): Indicates baseline DCT with Huffman coding.
• SOF9 (0xFFC9): Indicates extended sequential DCT with arithmetic coding.
• SOF10 (0xFFCA): Indicates progressive DCT with arithmetic coding.

A strict baseline decoder specifically checks for the SOF0 marker. When it encounters SOF9, SOF10, or other arithmetic variants, it rejects the file immediately as an unsupported format to prevent reading invalid data structures.

2. Missing DHT Markers and the Presence of DAC Markers

In a baseline JPEG, the decoder expects Define Huffman Table (DHT, 0xFFC4) markers to reconstruct the prefix code trees needed to parse the quantized discrete cosine transform (DCT) coefficients.

Arithmetic-coded streams do not contain DHT markers. Instead, they use Define Arithmetic Coding Conditioning (DAC, 0xFFCC) markers. These markers specify context binning, statistical conditioning, and parameters for the probability estimation machines. Baseline decoders do not have parsers for DAC segments. If a decoder attempts to continue parsing without finding the required DHT segments, it fails because it has no symbol tables to decode the entropy-coded segments.

3. Incompatible Entropy Decoding Engines

Even if a baseline decoder could bypass the marker checks, it lacks the execution engine required to process the compressed bitstream:

• Huffman Coding (Baseline): Maps discrete quantized coefficients to fixed-length or variable-length prefix code words. Decoding is a deterministic lookup-tree traversal.
• Arithmetic Coding (QM-Coder): Encodes the entire stream as a single fractional value within an interval between 0 and 1. It relies on a binary adaptive arithmetic coder known as the QM-coder. The QM-coder tracks dynamic intervals, performs bit-level renormalization, and constantly updates statistical probability models across multiple contexts.

Because the underlying data structures, registers, and state machines of a Huffman decoder cannot process interval-based probability states, feeding arithmetic-compressed bytes into a Huffman engine results in corrupted output, buffer overflows, or termination.

4. Historical and Licensing Causes

This functional divide exists because of intellectual property conditions present when the original JPEG specification was ratified in 1992. Arithmetic coding offered 5% to 15% better compression efficiency than Huffman coding, but it was heavily encumbered by patents held by IBM, AT&T, and Mitsubishi.

To ensure that JPEG could be freely and universally implemented without royalty burdens, the committee mandated Huffman coding for the baseline profile and relegated arithmetic coding to optional, extended profiles. Consequently, open-source reference implementations, such as early versions of the Independent JPEG Group’s libjpeg, disabled or completely omitted arithmetic coding logic. Decoders derived from these reference codebases remain strictly Huffman-based and reject arithmetic-coded streams by design.