identify-subject

name: identify_subject description: Identify a subject by matching a dental query image against the database.

Identify Subject Skill

Use this skill to find the identity of an unknown subject ("John Doe") by comparing their post-mortem (PM) dental X-ray against the database of ante-mortem (AM) records.

Dependencies

• DentalID.Core.Interfaces.IMatchingService (Vector Matching)
• DentalID.Core.Interfaces.IAiPipelineService (Feature Extraction)
• DentalID.Core.DTOs.DentalFingerprint (Biometric Data)

Workflow

1. Extract Features (Query):

   • Before matching, the query image MUST be processed to extract a FeatureVector.
   • Call IAiPipelineService.ExtractFeaturesAsync(stream).
   • Validation: Ensure FeatureVector is not null and has length 2048 (default embedding size).

2. Prepare Probe:

   • Create a DentalFingerprint object.
   • Set Code to "PROBE".
   • Set FeatureVector to the extracted float array.

3. Optimization: Smart Filtering:

   • Constraint: Matching against 10,000+ subjects is slow.
   • Action: Before calling FindMatches, filter the candidates list.
     • If AnalysisResult.EstimatedGender is "Male", exclude all "Female" subjects (allow "Unknown").
     • If EstimatedAge is 30, exclude subjects < 10 or > 80 (allow margin of error +/- 15 years).

4. Execute Match:

   • Call IMatchingService.FindMatches(probe, filteredCandidates).
   • This uses SIMD-accelerated Cosine Similarity.

5. Evaluate Results:

   • Matches are returned sorted by Score (0.0 to 1.0).
   • Thresholds:
     • > 0.80: High Probability Match.
     • 0.60 - 0.79: Possible Match (Requires manual review).
     • < 0.60: Inconclusive / No Match.

Example Usage (C#)

public async Task<List<MatchCandidate>> IdentifyUnknown(string queryImagePath, IAiPipelineService ai, IMatchingService matcher, IEnumerable<Subject> database)
{
    // 1. Extract
    using var stream = File.OpenRead(queryImagePath);
    var (vector, error) = await ai.ExtractFeaturesAsync(stream);
    
    if (vector == null) throw new Exception("Feature extraction failed");

    // 2. Probe
    var probe = new DentalFingerprint { FeatureVector = vector };

    // 3. Match
    var matches = matcher.FindMatches(probe, database);

    // 4. Report Top Match
    var best = matches.FirstOrDefault();
    if (best != null && best.Score > 0.85)
    {
        Console.WriteLine($"POSITIVE ID: {best.Subject.FullName} ({best.Score:P1})");
    }

    return matches;
}