Still Frame From Blurry Video: Plate, Face, Text

If the camera recorded enough pixels — yes: what usually gets in the way is blur, noise and a poorly chosen frame, not the pixel count. The tool finds the sharpest frame, stacks its neighbours and removes the blur. If the characters are three or four pixels tall, the information is physically absent and no software will recover it — looking for another camera is the honest answer.

Neural upscalers paint a plausible picture, and courts have already rejected such footage as evidence: the pixels are invented. Here the main processing is classical and reproducible: frame stacking, deconvolution, contrast equalisation. Neural reconstruction exists, but as a separate button, with a warning and a label on the result.

The noise in a recording changes from frame to frame, while the plate stays the same. Register 8–16 neighbouring frames on the object and average them: the noise cancels, the real detail remains. Registration follows the selected region, so it works on a moving car too, as long as it is seen at a similar angle.

Older cameras and DVRs write two moments in time into the even and odd lines of one frame, and store 720×576 with non-square pixels. Without field separation, sharpening turns the comb into a fence; without the pixel-aspect correction, circles become ovals and measurements lie. The tool detects interlacing itself and offers to separate the fields.

The tool does not replace an expert examination, but it provides what one values: the original frame next to the processed one, a list of steps with parameters that can be repeated, and the SHA-256 of the original file. All of it goes into a PDF report. Keep the original recording untouched — the processed frame is attached to it, not substituted for it.