A Private Web-Based Pipeline for Converting Images and PDF Pages into Editable PowerPoint Shapes

Abstract

Rasterized slide images and PDF pages are easy to distribute but difficult to reuse because text, layout objects, borders, and background regions are flattened into pixels. This paper presents a private web-based pipeline that converts images and PDF pages into editable PowerPoint files. The proposed system emphasizes local deployment, backend-centered image analysis, optical character recognition, text-region removal, geometric shape reconstruction, an intermediate Shape JSON representation, and .pptx export. The method first normalizes input pages, detects text regions, estimates background colors, extracts primary layout shapes through color clustering and contour analysis, and then reconstructs editable PowerPoint shapes with fill colors, borders, layer order, and placeholder text fields. A lightweight browser interface supports upload, preview, basic shape correction, and re-export, while computation remains on the local server. The current prototype completes the core conversion loop from raster input to editable PowerPoint output and provides a foundation for improved text removal, manual correction, batch PDF processing, and reusable private templates. The study demonstrates a practical engineering approach for transforming static visual documents into reusable presentation assets without relying on public cloud processing.