Prepare a document image
Correct a document image’s perspective using a manual quadrilateral crop, with optional sharpness and contrast adjustment and local PNG export.
How to use
Choose the image and wait for its preview. Enter corners in order: top-left, top-right, bottom-right, bottom-left. X and Y start at 0 at the top-left; X increases rightward and Y downward. Navigate with Tab and use Up/Down to change a value by 1. Review the crop boundary and set output width and height. Leave adjustment disabled or enable it and set sharpness and contrast, then correct perspective, review the PNG and download it.
For a 9×9 image, corners (0,0), (8,0), (6,8), (2,8) and a 5×5 output correct a trapezoid. The output center samples source position (4,16/3), rather than (4,4). Paper dimensions are not inferred; a different width-to-height ratio may stretch text and imagery.
Support & limits
Uses a real homography perspective transform with alpha-weighted bilinear interpolation. Crossed, concave, repeated, reversed, out-of-bounds and numerically unstable corners, or edges shorter than one pixel, are rejected; corners are never silently reordered or clipped. The preview is not a drag editor. Adjustment uses an alpha-weighted center/four-neighbor mean followed by contrast around 128, with channel rounding and clipping; it may amplify noise and does not recover missing detail. Interpolation and Canvas color handling may alter values even without adjustment; RGB at zero alpha becomes zero. Native processing works without OffscreenCanvas, and drawing may briefly delay the interface. Processing is local without uploads, automatic storage, camera capture, edge detection, OCR, AI or generated data. Preservation of original bytes, metadata, HDR or color profiles is not guaranteed; hidden data is not audited or securely redacted. The original stays unchanged and the result is a new PNG.
Common questions
Does this extract text or make a document searchable?
No. The output is only a PNG image. Arabic text already in the image remains pixels; no OCR, rewriting or invented words are provided. Review the output yourself, especially after perspective or color changes.