Agentic Image Content Extraction in Knowledge Base

5 min read

Overview

Image Content Extraction helps Unique AI understand information that appears inside images within PDF documents, such as charts, diagrams, infographics, scanned visual elements, and tables saved as images.

When this capability is enabled for a folder or space, Unique AI can extract a text description from visual content during ingestion. This makes important information from figures and charts easier to find through Knowledge Base search and available in AI chat answers.

For example, if a PDF contains a revenue chart, portfolio allocation diagram, or risk matrix, Image Content Extraction can help make that visual information searchable instead of relying only on the surrounding text.

Who is it for

Image Content Extraction is useful for users who work with documents where important information is often shown visually, such as:

• Financial Analysts reviewing annual reports, market updates, pitch books, and research documents
• Investment Teams working with portfolio allocation charts, performance graphs, and due diligence material
• Risk and Compliance Teams reviewing visual risk matrices, process diagrams, and regulatory reports
• Operations Teams searching through process diagrams, screenshots, and workflow documents
• Research Teams using reports that contain charts, diagrams, or image-based tables

If you do not see this capability or do not notice visual content being extracted, it may not be enabled for your folder, space, or organization. Please contact your administrator or internal support team.

Benefits

• Search visual information: Find information that appears in charts, figures, diagrams, and other images inside PDF files.
• Better answers from uploaded documents: AI chat can use more of the document content, including information that was previously only visible inside images.
• Improved understanding of image-heavy PDFs: Documents such as annual reports, financial presentations, and research reports become easier to analyze.
• No change to your upload workflow: You continue uploading files in the Knowledge Base or in chat as usual.
• More complete document processing: Text extracted from images is added to the processed document content, making it available for search and retrieval.

Use Cases

Financial Reports

Image Content Extraction can help make charts and figures searchable, such as:

• Revenue development charts
• Asset allocation graphics
• Performance dashboards
• Market trend visualizations
• KPI summaries shown as images

Research and Market Analysis

For research-heavy documents, it can help extract information from:

• Diagrams
• Infographics
• Competitive landscape visuals
• Scientific or market charts
• Tables embedded as screenshots or images

Risk and Compliance Documents

For risk and compliance teams, it can help with:

• Risk matrices
• Control diagrams
• Process flows
• Compliance dashboards
• Organizational charts

How It Works

When you upload a PDF, the document is processed by the Knowledge Base ingestion pipeline. If Image Content Extraction is enabled, the system checks whether the PDF contains figures or images that may include useful information.

For each detected figure, Unique AI tries to extract a text description or structured representation of the visual content. This extracted text is then added to the processed version of the document.

You do not need to manually mark images or charts. The process happens automatically during ingestion when the capability is enabled.

Step-by-Step Guide

Uploading Documents

Upload documents as you normally would:

1. Open the Knowledge Base.
2. Navigate to the folder where you want to upload the document.
3. Upload a PDF using the Upload Files button or drag and drop.
4. Wait for ingestion to complete.

Image Content Extraction only applies when the feature is enabled by your administrator for the relevant folder, space, or upload flow.

Checking Ingestion Status

After upload, check the ingestion state in the Knowledge Base:

• Queued for Ingestion: The file is waiting to be processed.
• Ingestion in Progress: The file is currently being processed.
• Ingestion Completed: The file has been processed and is ready for search and chat.
• Ingestion Failed: The file could not be processed successfully.

PDFs with many charts or figures may take longer to process than mostly text-based documents.

Searching Extracted Visual Content

Once ingestion is complete, you can search naturally for information that may have appeared inside charts or figures.

Example questions:

• “What does the revenue chart say about growth in 2024?”
• “Find documents showing asset allocation by region.”
• “Which report contains a risk matrix for operational risk?”
• “Summarize the diagram explaining the onboarding process.”
• “What KPIs are shown in the dashboard image?”

Search quality depends on the document quality, the visual clarity of the figures, and how the information is represented in the original PDF.

Using Extracted Visual Content in Chat

If the document is available to a chat space, the assistant can use the extracted visual content when answering questions.

For example, a user may ask:

• “What trend is shown in the chart on revenue development?”
• “Summarize the main figures in this report.”
• “Which charts mention Switzerland?”
• “What does the portfolio allocation graphic show?”

The assistant can then retrieve and use the extracted content like other ingested document text.

What To Expect

Processing Time

Image Content Extraction can increase processing time because each relevant figure may require additional analysis. Documents with many charts, diagrams, or image-heavy pages may take noticeably longer to ingest.

Search Results

After successful ingestion, visual content may appear in search results or be used by AI chat. This does not change the original file. It only improves the processed searchable content created from the file.

Existing Documents

Documents that were uploaded before Image Content Extraction was enabled may need to be reprocessed or uploaded again before their visual content becomes searchable. If you are unsure whether a file has been processed with this capability, contact your administrator.

Best Practices

• Use PDFs with clear, high-quality charts and images.
• Prefer documents where figures include readable labels, legends, and captions.
• Avoid very low-resolution scans when possible.
• For important documents, check whether ingestion completed successfully before using them in chat.
• If expected chart information is missing, try asking a more specific question or contact your administrator.

Limitations

Image Content Extraction improves access to visual information, but it is not perfect.

• Very small or blurry figures may not be extracted reliably.
• Complex charts may be summarized rather than fully converted into exact data tables.
• Some numerical values may be difficult to read if they are not clearly visible.
• Handwritten content may not be recognized reliably.
• Decorative images, icons, and logos may be skipped or only briefly described.
• Some text may appear twice if it was already detected by standard document processing and also extracted from a figure.
• The feature currently focuses on visual content inside PDF processing.

Frequently Asked Questions

Do I need to enable anything myself?

Usually no. Your administrator controls whether Image Content Extraction is enabled for a folder, space, or tenant. You continue uploading and searching documents as usual.

Why do some PDFs take longer to process?

PDFs with many charts, figures, or image-heavy pages may require additional processing. This can increase ingestion time.

Does this change the original document?

No. The original file is not changed. Unique AI creates searchable processed content from the file during ingestion.

Will every chart be extracted perfectly?

No. Extraction quality depends on the image quality, chart complexity, resolution, and how clearly the information is shown.

Can I use it for documents already uploaded?

Existing documents may need to be reprocessed or uploaded again before Image Content Extraction is applied. Contact your administrator if you need this for existing content.

What should I do if visual information is missing?

First confirm that ingestion completed successfully. If the document is important and the visual content is still missing, contact your administrator or internal support team.