Azure OCR

Use this property if connecting to a self-hosted instance of Azure Cognitive Services.

When not set, the cloud instance of Azure Cognitive Services will be used (https://{ApiRegion}.api.cognitive.microsoft.com/vision), and an API Key will be required.

Need to try this feature out? At the time of this writing, Microsoft provides 5,000 free calls per month when a new account is created. New Azure accounts can be created here.

Can be one of the following values:

Select the region nearest the location where Grooper servers will be running.

Can be one of the following types:

Can be one of the following types:

Can be one of the following types:

If no value is provided, multi-language OCR will be enabled. Please note that Azure does not detect the language of documents.
Run the Detect Language activity after extraction to ensure the proper culture code is assigned to the document.

Retries are triggered for network timeouts, service rate limiting (e.g., "too many requests"), or other transient errors encountered when communicating with the Azure endpoint.

• If a recoverable error occurs, the engine will wait 1 second and then retry, up to the specified number of times.
• If set to 0, no retries will be attempted and the first error will be thrown immediately.
• Increasing this value can improve reliability in unstable network environments or when processing large batches, but may increase total processing time if repeated failures occur.

Best Practice:
Use the default value (2) for most scenarios. Increase only if you experience frequent transient errors and are willing to tolerate longer processing times for improved robustness.

• Value range: 0.0 (no filtering) to 1.0 (only perfect confidence).
• Words with a confidence below this threshold are excluded from the final results.
• Confidence is determined by the Azure OCR engine for each recognized word.

Best Practice:
Start with a value around 0.5 to filter out low-confidence words, and adjust as needed based on your document quality and desired accuracy. Setting this too high may remove valid but slightly uncertain words; setting it too low may allow more errors.

• Value range: 0.0 (no filtering) to 1.0 (only perfect confidence).
• Segments (lines) with a confidence below this threshold are excluded from the output.
• Segment confidence is based on the aggregate confidence of the words in the line, as reported by Azure.

Best Practice:
Use a moderate value (e.g., 0.5 to 0.8) to filter out unreliable lines, especially in noisy or low-quality scans. Increase the threshold to improve accuracy, or decrease it to maximize recall.

When enabled, any text segment identified as handwriting by Azure will be filtered out during post-processing.

• This is useful for forms or documents where only machine print is desired, or where handwriting is likely to be misread.
• Handwriting segments are still detected and validated, but are excluded from the final results if this property is true.
• The validation logic (see IsHandwritingMisread) further ensures that handwriting is suppressed only when it overlaps with high-confidence traditional OCR segments, reducing the risk of discarding valid data.

Best Practice:
Enable this option if your documents contain handwriting that should not be extracted, or if you observe frequent misreads in handwritten regions.

If set, this profile determines the configuration of the traditional OCR engine (such as Transym) used in parallel with Azure OCR.

• The traditional OCR results are used to validate, supplement, and repair Azure results, and to break ties when the two engines disagree.
• If not set, a default Transym OCR configuration is used.
• You can use this property to fine-tune traditional OCR settings (e.g., language, segmentation, filtering) for your specific documents.

If set, this lexicon is used to break ties in cases where Azure and Transym disagree on the value of a text segment.

• When both engines produce different results for a word or line, the engine will prefer the value found in the vocabulary.
• This is especially useful for domain-specific terms, abbreviations, or uncommon words that may be misread by OCR engines.
• If not set, only confidence scores and preferred patterns are used for tie-breaking.

Best Practice:
Provide a custom lexicon if your documents contain specialized terminology, names, or values that are not common in general language models.

This regular expression should define patterns which occur frequently in the document set and are important for extraction accuracy.

• When Azure and traditional OCR disagree, the engine will prefer the value that matches one of these patterns.
• Use the regular expression 'OR' operator (|) to include multiple patterns.
• Patterns should be tailored to the types of data you expect (e.g., invoice numbers, dates, currency, IDs).

Example:
For invoices, to prefer currency values: $?\d{1,3}(,?\d{3})*.\d{2}

Best Practice:
Define patterns for the most critical or error-prone fields in your documents to maximize extraction reliability.