PDF/UA-2

ISO 14289-2 (PDF/UA-2) significantly improves the accessibility of PDF. Foxit's PoC implementation of various PDF creation and consumption tools

We provide Installer for our PoC implementations. The PDF Sample files we produced to achieve interoperability.

We are testing several AI approaches to help with automatic math recognition as a part of remediation processes. The results of our discoveries can be found here: Recognition

If you are interested in interoperable testing, have comments, or want to report a bug or suggestion, please use the Issues feature on GitHub.

PDF 2.0 and PDF/UA-2 features

The ISO 32000-2:2020 (PDF 2.0) specification only provides file format requirements. It should give clear guidance for PDF creation tools. What should and should not be present in the compliant file, but it is not clear how the processor should use all that information. We use this GitHub repository to explain our reasoning with implementations and sample files so we can reach a consensus with other developers. It is essential to give end users the same experience when consuming PDF UA-2 files through AT tools.

The following sections explain our implementation decisions. Feel free to comment.

MathML

PDF 2.0 in conjunction with PDF/UA-2 recognizes two methods of semantically denote math. Both methods require the use of Formula structure element and then:

• Use <mathml> elements in the MathML namespace as direct children
• or the associated MathML data in the AF property of the structure element

In our PoC we use the second option.

Logic in providing Associated file

PDF 2.0 doesn't provide any guidance regarding processors, so it's not specified how a processor should deal with specific scenarios.

Our algorithm when processing PDF files with AF is as follows:

For Formula structure element:

1. Search Criteria:

   • Focus only on AFs where AFRelationship == Supplement (as per PDF/UA-2).
   • Ensure that the Mediatype (Subtype key) is application/mathml+xml.
     • Note: The Mediatype comparison is currently a string match, with ongoing discussions in the community about potential variations in the key.

2. AF Structure Handling:

   • Accept both arrays and directories for AF. While PDF 2.0 officially allows only arrays, some files use AF as a dictionary.
   • • Only the first element that meets the criteria is used

3. Fallback Mechanism:

   • If no suitable AF entry is found proceed to process the children (kids) of the structure element.
   • If Formula doesn’t have NS == 2.0 then the old processing is used.

4. Unimplemented Features:

   • AFRelationship == Alternative is not yet implemented. The usage of this key is currently under discussion.

5. Priority Rules:

   • AF has priority over Alt.
   • Alt has priority over ActualText

1. Substructure Handling:

   • If a suitable AF is found, process the substructure of the element and provide the content of the stream data as the textual value, i.e. only that content is replaced, not the substructure.
   • If ActualText is present we ignore AFs and substructure. ActualText serves as a full replacement of structure element (this is true for all structure elements. Formula is no different)

2. Processing Formula Pseudocode

IF AF is present AND Namespace == 2.0 AND AFRelationship == Supplement AND Mediatype == application/mathml+xml THEN 
   // Only the first element that meets the criteria is used
   // Process the substructure and provide the stream data content as textual value
   Provide stream data content, replacing content but not substructure
   
ELSE IF Alt is present THEN
    // Provide Alt
    Provide Alt

ELSE IF ActualText is present THEN
   // Ignore AFs and substructures, and provide ActualText
   Provide ActualText

ELSE
    // Process the substructure of the element
    Provide content and process substructure

Rolemap and RolemapNS Processing

1. General Rolemap Handling:

   • Process the Rolemap or RolemapNS on a Structure Element (SE) in a PDF, regardless of the PDF version.

2. Namespace (NS) Check:

   • If the SE does not have a Namespace (NS), check the Rolemap.
   • If the SE does have an NS and it is other than PDF 1.7, PDF 2.0, or MathML, check RolemapNS.

3. Custom Element Mapping:

   • Allow mapping of a single custom element into two different standard elements based on RolemapNS.

4. Backward Compatibility:

   • If the Formula element does not have NS == 2.0, then revert to the old processing method. This is our backward-compatible hack since older processors won’t check the NS.

5. Namespace Flexibility:

   • Elements could be in any namespace (e.g., LaTeX). The discussion of NS == 2.0 occurs after rolemapping (i.e., RolemapNS).