Visual

Classifies documents or pages based on their visual appearance using image-based features, without requiring OCR.

Remarks

The Visual classification method assigns Document Types by analyzing the visual characteristics of document images, rather than relying on text extraction or content rules.

Overview

• Visual classification uses an IP Profile to extract image-based features from each page or document.
• Features may include layout, graphics, logos, line structures, or other visual patterns detected by the configured IP Profile.
• This method does not require OCR, making it suitable for image-only documents, poor-quality scans, or real-time classification during scanning.

How It Works

1. For each document or page, the configured IP Profile is applied to extract a set of visual features.
2. These features are compared to trained models for each Document Type, measuring visual similarity.
3. The Document Type with the highest similarity score is assigned to the document or page.
4. Classification can be performed at the document or page level, supporting both single- and multi-page scenarios.

Configuration Guidance

• Assign an IP Profile that includes an 'Extract Features' command tailored to the visual elements most distinctive for your document types.
• Train each Document Type with representative samples to build accurate visual models.
• Use in environments where text extraction is unreliable, unavailable, or unnecessary.
• Combine with other classification methods (such as Rules-Based or Lexical) for hybrid solutions, if needed.

When to Use

• Ideal for classifying forms, templates, or documents with consistent visual layouts but variable or missing text.
• Useful for real-time document separation and classification during scanning, where speed and independence from OCR are required.
• Effective for image-only documents, such as faxes, checks, or forms with graphical elements.

Practical Notes

• The accuracy of visual classification depends on the quality and consistency of the images and the effectiveness of the IP Profile.
• No OCR is performed, so text content is ignored; only visual features are considered.
• Ensure the IP Profile contains an 'Extract Features' command, or validation will fail.
• Visual models must be trained for each Document Type to enable accurate classification.

For more information, see the documentation for IP Profile, Document Type, and training procedures for visual classification.

Properties

Name	Type		Description
Ip Profile	IP Profile	►	The IP Profile to be used for extracting image-based features. EditorNode Reference Editor Requiredtrue The 'IP Profile' property specifies which IP Profile is applied to each document or page to extract visual features for classification. What is an IP Profile? An IP Profile defines a sequence of image processing operations—such as filtering, edge detection, or feature extraction—to be performed on an image. It can include multiple IP Step and IP Group objects, allowing for both simple linear sequences and complex, conditional processing flows. The execution order of steps is determined by the profile's structure and any flow control expressions defined on groups or steps. Required IP Commands for Visual Classification The IP Profile must include an Extract Features IP Command. This command is responsible for generating the set of visual features used for classification. Additional IP commands (such as deskew, despeckle, or other cleanup steps) may be included before 'Extract Features' to improve feature quality, but are optional. How It Is Used in Visual Classification During classification, the selected IP Profile is executed on each image to generate a set of visual features. These features are then compared to trained models for each Document Type to determine the best match. Only the features produced by the 'Extract Features' command are used for classification; other commands serve to prepare the image as needed. Configuration Guidance Choose or design an IP Profile that highlights the most distinctive visual elements of your document types (such as layout, logos, or graphical patterns). Always include an 'Extract Features' step as the final or one of the final steps in the profile. You can reuse IP Profiles across multiple classification methods or activities, or create specialized profiles for different document sets. Practical Notes The quality and relevance of the extracted features depend on the configuration of the IP Profile and the placement of the 'Extract Features' command. Modifying the IP Profile may require retraining visual models for each Document Type to maintain classification accuracy. For more information on creating and managing IP Profiles, see the documentation for IP Profile, IP Step, IP Group, and the 'Extract Features' IP Command.

See Also

IP Profile

Used By

Content Model