Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Scope and Application
“content”: “
The standard IEC 14495-1-02, technically identical to ISO/IEC 14495-1:1999 together with its Amendment 1:2002, specifies the baseline algorithm for lossless and near-lossless compression of continuous-tone still images. Commonly known as JPEG-LS, this standard is also adopted in Canada as CAN/CSA-ISO/IEC 14495-1-02. It provides a low-complexity, efficient coding method suited for applications where no loss or only a controlled, bounded loss of information is acceptable, such as medical imaging, remote sensing, and digital archival.
” “Scope and Application
IEC 14495-1-02 defines a compression algorithm for continuous-tone (multi-level) still images. It is explicitly not intended for bi-level or palletized images. The standard covers two modes:
- Lossless mode — Perfect reconstruction of the original image samples.
- Near-lossless mode — Reconstruction with a controlled maximum absolute error specified by the parameter
NEAR.
The algorithm is designed for low-complexity hardware and software implementations, offering compression ratios competitive with more complex schemes (e.g., JPEG lossless) while requiring significantly less computation. Typical applications include:
- Medical imaging (DICOM compliant)
- Geospatial and satellite imagery
- Digital photography archival
- Industrial inspection
Tip: For medical imaging systems requiring reversible compression, JPEG-LS remains a popular choice due to its deterministic, lossless behavior and low latency.
Technical Requirements
The baseline algorithm is based on a predictive coding scheme with context modeling and Golomb–Rice entropy coding. The key technical elements are:
Prediction and Context Modeling
Each pixel is predicted using a fixed, edge-detecting predictor (the “LOCO-I” predictor) that examines three previously coded neighbors. A local gradient vector is computed and quantized into one of 365 contexts, which adaptively selects the Golomb–Rice parameter and sign model.
Coding Modes
JPEG-LS switches between regular mode and run-length mode:
- Regular mode — used for non‑uniform image regions; it encodes prediction residuals using Golomb–Rice codes optimized per context.
- Run-length mode — used for flat regions (identical pixel values), encoding consecutive replicates efficiently.
Key Parameters
| Parameter | Symbol | Allowed Values | Description |
|---|---|---|---|
| Maximum sample value | MAXVAL | 2, 3, …, 216–1 | Defines the dynamic range of image samples (typically 255 for 8‑bit images). |
| Near-lossless bound | NEAR | 0, 1, …, MAXVAL/2 | Maximum absolute reconstruction error. NEAR = 0 gives lossless. |
| Run-length reset threshold | RESET | 1, …, 255 | Controls how often Golomb–Rice parameter is reset in run mode. |
| Number of points in threshold table | NST | 1, …, 255 | Number of quantization thresholds for context gradients. |
Warning: The near-lossless mode (
NEAR > 0) introduces a deterministic, bounded error. However, repeated encode/decode cycles can accumulate errors unless the same parameters are used throughout. Implementation Highlights
The standard mandates a baseline that every compliant encoder and decoder must support. Extensions (e.g., for more color components, higher bit depths) are defined in separate documents. The encoder architecture is straightforward:
- Compute prediction and local gradients.
- Determine context and coding mode (regular or run-length).
- Encode residuals (regular) or run counts (run‑length) using Golomb–Rice codes.
- Update context models adaptively.
Decoding is symmetric and equally low in complexity. This simplicity allows real‑time decompression even on embedded platforms.
Image Dimensions and Scan Order
Images are assumed to be rectangular, with up to 231–1 pixels per dimension. The standard specifies raster‑scan order (left to right, top to bottom) and multiple‑component interleaving options (e.g., pixel‑interleaved “LINE” or “SAMPLE” interleaving).
Good practice: For maximum compression efficiency, adjust the
NEAR parameter to the smallest value that still satisfies the application’s quality requirements. Even a NEAR of 1 can significantly increase compression ratios while remaining visually transparent for many images. Compliance and Testing
Conformance to IEC 14495-1-02 is assessed through test procedures defined in ISO/IEC 14495-2 (JPEG-LS Part 2: Compliance testing). Testing covers:
- Verification that encoded bitstreams conform to the syntax and coding rules.
- Decoding of reference test streams and comparison with known output.
- Checking of all error‑resilience features (if implemented).
Manufacturers of medical imaging equipment, for example, must demonstrate that their JPEG-LS implementation passes the conformance tests to claim compliance with DICOM’s transfer syntax for lossless compression.
Important: Non‑compliant encoders may generate bitstreams that, while decodable, do not guarantee the advertised reconstruction bounds (especially in near‑lossless mode). Always use certified implementations when legal or quality requirements demand strict adherence.
Adoption and References
The Canadian version CAN/CSA-ISO/IEC 14495-1-02 is an identical adoption, containing no deviations. Other national standards bodies have similarly adopted the text of ISO/IEC 14495-1:1999 with Amendment 1:2002.
Frequently Asked Questions
Q: What is the main difference between JPEG-LS and standard JPEG (ISO/IEC 10918-1)?
A: JPEG-LS is optimized for lossless/near‑lossless compression using a low‑complexity predictive scheme, while baseline JPEG is primarily lossy and uses DCT transforms. JPEG-LS typically achieves better compression ratios for lossless coding than JPEG lossless mode.
A: JPEG-LS is optimized for lossless/near‑lossless compression using a low‑complexity predictive scheme, while baseline JPEG is primarily lossy and uses DCT transforms. JPEG-LS typically achieves better compression ratios for lossless coding than JPEG lossless mode.
Q: Can IEC 14495-1-02 be used for color images?
A: Yes, but the baseline algorithm processes each component independently (e.g., RGB can be compressed as three separate planes). For inter‑component decorrelation, the standard allows choosing a color transformation (e.g., reversible color transform) as an extension, though not part of the baseline.
A: Yes, but the baseline algorithm processes each component independently (e.g., RGB can be compressed as three separate planes). For inter‑component decorrelation, the standard allows choosing a color transformation (e.g., reversible color transform) as an extension, though not part of the baseline.
Q: Is JPEG-LS suitable for video compression?
A: JPEG-LS is a still‑image compression standard. For video, it may be used to compress individual frames (motion‑JPEG‑LS), but it does not exploit temporal redundancies. Dedicated video codecs are more efficient for moving imagery.
A: JPEG-LS is a still‑image compression standard. For video, it may be used to compress individual frames (motion‑JPEG‑LS), but it does not exploit temporal redundancies. Dedicated video codecs are more efficient for moving imagery.
Q: Do I need a license to implement JPEG-LS?
A: JPEG-LS was developed with the aim of being “royalty‑free.” However, implementers should review the most current patent pool status. As of 2026, no licensing fees are typically required for compliant implementations.
A: JPEG-LS was developed with the aim of being “royalty‑free.” However, implementers should review the most current patent pool status. As of 2026, no licensing fees are typically required for compliant implementations.
Article prepared in accordance with ISO/IEC Directives Part 2. Standard references: IEC 14495-1-02 (ISO/IEC 14495-1:1999/Amd 1:2002), CAN/CSA-ISO/IEC 14495-1-02. — 2026
“