Enhance Scanned PDF Online Free

Scanned PDFs lose quality for reasons that have nothing to do with the document itself. Flatbed and sheet-fed scanner sensors introduce grain, especially at lower DPI settings or in low light. Phone-camera scanning apps add uneven lighting, shadows near the spine of a book, and a slight perspective tilt. Photocopying a document more than once compounds contrast loss and noise each generation. Old paper yellows, and cheap toner or faded ink drops contrast between text and background. None of this is a flaw in the original document — it is signal degradation introduced entirely by the capture process, and most of it is measurable and correctable.

PDFcrest's enhancer addresses this with nine ordered stages. First, if the page is tilted, PDFcrest's projection-profile deskew detector (the same one used in the standalone Deskew PDF tool) straightens it, so every later stage operates on an upright page. Next, a background-flattening step estimates the paper's own local background color per color channel using a large-radius box blur, then divides each pixel by that local background and rescales toward white — this is what actually removes yellowing, scanner shadows, and uneven lighting, because a warm-toned background gets pulled toward neutral independently per channel rather than just brightened uniformly. With a clean, even background established, a brightness stage reads the page's own 90th-percentile brightness (a proxy for how bright its paper background already is, since background pixels dominate a typical text page) and solves a gamma correction that pushes that level toward near-white — targeting the background specifically, rather than the whole page's average tone, avoids dragging text-heavy pages toward mid-gray. Then CLAHE (Contrast Limited Adaptive Histogram Equalization) enhances local contrast in tiles across the page rather than applying one global curve — this is what separates faded ink from background even when different regions of the same page were scanned at different exposures. Finally, an edge-aware noise reduction pass smooths grain while preserving character edges, followed by two sharpening passes (a fine unsharp mask for character edges and a coarser one for overall text clarity) and an optional black-level clip and black & white threshold for maximum OCR contrast.

The mathematics behind a few of these stages are genuinely simple. Auto-gamma brightness solves for the exponent γ in output = 255 × (input/255)^γ such that the page's average brightness lands on a target mid-tone — γ below 1 brightens a dark page, γ above 1 darkens an overexposed one. CLAHE's contrast stretch within each tile follows the classic histogram-equalization formula s = (L-1) × CDF(r), where a pixel's new value s comes from the cumulative distribution of gray levels in its local tile rather than the whole page, then a "clip limit" caps how much any single gray level can be boosted so noise in flat regions doesn't get amplified into visible speckle. None of this requires deep learning — it's the same category of classical image-processing math that has underpinned document scanning software for decades, applied consistently and combined into one pipeline.