ComPDFKit Conversion SDK V3.0 is officially released. This version has a full-stack technology upgrade, including AI layout analysis, AI table recognition, and a self-developed algorithm to restore natural reading order. These upgrades greatly improve the accuracy, layout restoration, and speed of PDF conversion.
To test the real effects of these improvements, we used the PDF-to-Word feature as an example. We checked how well the layout was restored, how accurate the font and text conversion was, and how tables and images were handled. Below is a detailed test report showing the major progress and performance of Conversion SDK V3.0 in PDF conversion.
Test Preparation for ComPDFKit Conversion SDK V3.0
Before starting the testing, we did the following preparations to ensure the accuracy and validity of the test results.
1. Testing Sample Documents
We had chosen 30 PDF sample documents of different sizes and structures. What's more, all selected documents met the following requirements to evaluate the SDK’s stability and compatibility with basic features:
- No OCR content or scanned images.
- No complex content with mixed images and texts, mathematical formulas, or nested tables spanning across pages.
- Only contains merged cells and no nested tables within tables.
- Embedded images with a moderate quantity and no floating layouts.
- The content mainly included standard office elements, such as body text, annotations, lists, headers & footers, and hyperlinks.
2. Conversion Parameters Settings
In this testing, we applied the same conversion parameters to ensure the consistency and comparability of the test results. The specific configurations are as follows:
PDF Page Layout Testing
For testing the accuracy of page layout recovery during PDF conversion, we focused on four key layout elements: margins, multi-column layout, headers, and footers.
Margins were restored with an accuracy of 86%. The original page boundaries and layout were mostly kept, providing a solid base for further editing in Word.
The multi-column layout was restored with an accuracy of 60%. The main issue was changes in tracking and line spacing after converting the PDF to Word, which caused differences in the number of lines between columns. This affected the layout’s symmetry and balance.
The header and footer accuracy was below 50%, similar to other tools on the market. Because of their varied structure and style, this feature is still basic. Nevertheless, it is already able to recognize more than 90% of document content. For more complex documents, we provide customized intelligent document processing solutions to meet specific needs.
PDF Font & Text Testing
We tested the text conversion results of ComPDFKit Conversion SDK V3.0 during the PDF-to-Word process, focusing on three key areas: font properties, text structure, and paragraph formatting accuracy.
For sentence structure and paragraph restoration, the accuracy in splitting paragraphs and avoiding missing text was over 70%. More impressively, the accuracy of separating paragraphs was 90%, which preserved the original meaning and flow of the content. In terms of paragraph formatting, such as text alignment, line spacing, and space before and after paragraphs, the accuracy was around 80% to ensure the conversion results keep the original layout and avoid text jumping page.
The SDK excelled in recognizing font attributes, with 100% accuracy for restoring font color, bold, strikethrough, and scaling. Font size and italic styles were recognized with over 90% accuracy, showing strong compatibility and stability in handling complex text styles.
PDF Table & Image Testing
We test the image and table element rendering capabilities of ComPDFKit Conversion SDK V3.0 in PDF-to-Word conversion.
For image rendering, the Conversion SDK V3.0 performed steadily across 30 sample documents, with an accuracy rate of over 80% for image size, position, transparency, and text wrapping. In most documents, the image styles and layouts were well preserved when converted to Word. Ensure a good visual consistency and compatibility.
In PDF conversion with tables, the overall accuracy remained above 60%. The accuracy for table position restoration exceeded 90%, and the recognition accuracy for merged cells reached 80%. Our Conversion SDK V3.0 shows strong adaptability for the most common table structures. Besides, the restoration of style properties like cell background color was also stable, with conversion results outperforming traditional tools in both structural integrity and style retention.
Performance Testing
Besides evaluating the layout and content accuracy of PDF to Word conversion, we also tested the performance of ComPDFKit Conversion SDK V3.0. covers real-world scenarios with different file sizes and page numbers to verify its stability and processing efficiency under high load.
According to the test results, the average conversion speed of ComPDFKit Conversion SDK V3.0 remained stable at 0.5–0.8 seconds per page. With high accuracy in rendering maintained, it demonstrated good performance, meeting the demands of large-volume or high-frequency conversions.
Conclusions
Based on the test results, ComPDFKit Conversion SDK V3.0 performs well in core conversion accuracy, layout restoration, and PDF to Word processing speed. By using AI layout analysis and our self-developed algorithms, it greatly improves the ability to handle complex documents, especially in key details like margins, fonts, images, and tables. Although there is still room for improvement in some areas, it can already meet over 90% of document recognition needs.
In addition, for more complex documents, we offer customized intelligent document processing solutions to use an AI solution to ensure 98% accuracy. Help grow your business.
In the future, ComPDF will continue to drive technological innovation, improve user experience, and strive to provide more accurate and efficient document conversion solutions.