Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
Understanding IEC 13249-5-04: SQL Multimedia and Application Packages – Part 5: Still Image
A Technical Guide to Still Image Data Management Using SQL/MM
Scope and Purpose
IEC 13249-5-04 (identical to CAN/CSA-ISO/IEC 13249-5-04) is part of the ISO/IEC 13249 series that extends the SQL standard with multimedia and application-specific packages. Part 5 specifically addresses still image data. The standard defines a set of SQL user‑defined types (UDTs), routines, and schemas that enable the storage, management, and retrieval of still images based on extracted visual features. It provides a standardized way to perform content‑based image retrieval and similarity comparison directly within SQL queries, without requiring external image analysis tools.
Standardization benefits: IEC 13249-5-04 promotes interoperability across database systems, reduces development effort for multimedia applications, and ensures a consistent programming model for image data management.
Technical Architecture
User‑Defined Types (UDTs) and Schema
The standard defines several UDTs that are organized under the schema SQLMM_STILLIMAGE. Each UDT encapsulates image content and/or a specific visual feature. The following table summarizes the primary UDTs:
| UDT | Purpose | Key Attributes or Methods |
|---|---|---|
SI_StillImage | Represents a still image with content and optional feature data. | imageContent, imageFormat, imageType, height, width |
SI_AverageColor | Captures the average color of an image region. | colorSpace, red, green, blue, alpha |
SI_ColorHistogram | Describes the global colour distribution via a histogram. | colorSpace, bins, binValues |
SI_PositionalColor | Records colour at a specific spatial location. | xPosition, yPosition, colorValue |
SI_Texture | Encodes textural properties (coarseness, contrast, directionality). | coarseness, contrast, directionality |
SI_EdgeHistogram | Captures edge orientation distribution. | edgeAngles, frequencyArray |
Each UDT includes constructor functions, observer methods, and comparison routines. For example, an SI_StillImage object can be created from a BLOB or a file locator, and its features can be extracted using methods such as SI_StillImage::getAverageColor() and SI_StillImage::getColorHistogram().
Feature Extraction and Similarity Comparison
The standard provides a set of SQL routines to compute visual features from raw image data. These routines include SI_GetAverageColor, SI_GetColorHistogram, SI_GetTexture, and SI_GetEdgeHistogram. The extracted feature objects expose a similarity() method that returns a score (typically between 0 and 1) indicating the likeness between two feature instances.
Example query pattern:
SELECT image_id, image_path FROM images WHERE image_data.SI_StillImage::getColorHistogram() .similarity(SI_ColorHistogram(:queryHistogram)) > 0.85 ORDER BY image_data.SI_StillImage::getColorHistogram() .similarity(SI_ColorHistogram(:queryHistogram)) DESC; Implementation tip: Pre‑extract features and store them in separate columns or materialized views to avoid repeated computation, especially when dealing with large image volumes.
Image Storage and Encoding
The standard supports common image formats such as JPEG, PNG, GIF, and TIFF. Image data can be stored directly as BLOB columns or referenced external files through SQL locators. The UDT SI_StillImage includes attributes for image format and type, allowing the DBMS to handle decoding and feature extraction transparently.
Implementation Considerations
Performance Tuning
Feature extraction is computationally intensive. For large databases, it is advisable to compute features only once and store them alongside the images. Consider using partitioned tables and parallel index scans to accelerate similarity queries.
Performance warning: Without appropriate indexing (e.g., R‑trees or specialized spatial indexes on feature vectors), exhaustive scanning of all rows can make similarity queries impractically slow.
Integration with Existing SQL
The UDTs and routines defined by IEC 13249-5-04 are designed to coexist with standard SQL data types. They can be used in SELECT, INSERT, UPDATE, and DELETE statements as well as in table creation and indexing. The standard also specifies conformance levels, allowing vendors to implement subsets.
Compliance and Certification
Conformance to IEC 13249-5-04 requires implementation of all mandatory SQL statements, UDTs, and routines defined in the standard. Vendors may claim full conformance or may declare partial conformance if they implement only a subset of features (e.g., only colour‑based comparison without texture).
Testing suites for conformance are available through ISO/IEC published test assertions. Adoption of the standard remains limited; however, the architectural principles have influenced later SQL‑based multimedia extensions in several commercial databases.
Common pitfall: Relying solely on a single feature type (e.g., only colour) for high‑precision queries can yield poor results. Always combine multiple features and calibrate similarity thresholds based on empirical testing for your specific image corpus.
Frequently Asked Questions
Q: How does IEC 13249-5-04 relate to other parts of SQL/MM?
A: Each part of SQL/MM (Full‑Text, Spatial, Still Image, Data Mining, etc.) is independent but shares common foundations. They can be implemented together in a single SQL environment, and the design patterns for UDTs and routines are consistent across parts.
A: Each part of SQL/MM (Full‑Text, Spatial, Still Image, Data Mining, etc.) is independent but shares common foundations. They can be implemented together in a single SQL environment, and the design patterns for UDTs and routines are consistent across parts.
Q: Is IEC 13249-5-04 still current, and what is the latest edition?
A: The original publication was ISO/IEC 13249-5:2003 (which IEC 13249-5-04 mirrors). A revised edition, ISO/IEC 13249-5:2011, added new features including scaling methods and improved metadata handling. Check with your DBMS vendor for which edition they support.
A: The original publication was ISO/IEC 13249-5:2003 (which IEC 13249-5-04 mirrors). A revised edition, ISO/IEC 13249-5:2011, added new features including scaling methods and improved metadata handling. Check with your DBMS vendor for which edition they support.
Q: Which database systems implement SQL/MM Still Image?
A: IBM DB2 and Oracle have historically provided support for SQL/MM multimedia types, though full compliance is not widespread. Many systems now offer proprietary multimedia extensions. Always consult vendor documentation for conformance details.
A: IBM DB2 and Oracle have historically provided support for SQL/MM multimedia types, though full compliance is not widespread. Many systems now offer proprietary multimedia extensions. Always consult vendor documentation for conformance details.
Q: What are the main challenges when deploying IEC 13249-5-04?
A: Performance of similarity queries, handling of different image formats, and the need to tune feature thresholds are the most common challenges. Proper indexing and pre‑computed feature tables are strongly recommended.
A: Performance of similarity queries, handling of different image formats, and the need to tune feature thresholds are the most common challenges. Proper indexing and pre‑computed feature tables are strongly recommended.