Correct Video Perspective Distortion with FFmpeg

This article demonstrates how to correct perspective distortion—often caused by camera tilt, angled shooting, or keystoning—using the powerful perspective video filter in FFmpeg. You will learn the underlying coordinate mapping system, the syntax of the filter, and a step-by-step practical example to flatten and correct your video’s perspective.

Understanding the FFmpeg Perspective Filter

The perspective filter works by mapping four corner points of a distorted quadrilateral in the input video to four new corner points in the output video. By defining where the tilted corners currently are and where they should be, FFmpeg warps the video frame to correct the distortion.

The four corners are defined in a specific order: 1. Top-Left (0) 2. Top-Right (1) 3. Bottom-Left (2) 4. Bottom-Right (3)

The Filter Syntax

The basic syntax for the perspective filter is:

perspective=x0:y0:x1:y1:x2:y2:x3:y3:px0:py0:px1:py1:px2:py2:px3:py3:sense

x0 to y3: The \(X\) and \(Y\) coordinates of the four corners in the original, distorted input video.
px0 to py3: The target \(X\) and \(Y\) coordinates of the four corners in the corrected output video.
sense (Optional): Specifies the conversion action. Use 0 (default) for mapping destination to source, or 1 for mapping source to destination.

Step-by-Step Correction Guide

Step 1: Identify the Distorted Coordinates

Open your video in a media player or editing software that shows pixel coordinates (such as Avidemux, Photoshop, or even using FFmpeg’s ffplay with the drawgrid filter).

Locate the four corners of the distorted object or frame boundary you want to correct: * Top-Left (x0, y0): e.g., (150, 80) * Top-Right (x1, y1): e.g., (1130, 120) * Bottom-Left (x2, y2): e.g., (80, 980) * Bottom-Right (x3, y3): e.g., (1200, 900)

Step 2: Define the Target Coordinates

Decide where these corners should map to in the final output to look perfectly flat. If you want the corrected area to fill a standard 1280x720 frame, your target coordinates will be the outer boundaries of the video: * Top-Left (px0, py0): (0, 0) * Top-Right (px1, py1): (1280, 0) * Bottom-Left (px2, py2): (0, 720) * Bottom-Right (px3, py3): (1280, 720)

Step 3: Run the FFmpeg Command

Combine these coordinates into the FFmpeg command. Below is an example command applying the perspective filter and copying the audio stream without re-encoding it:

ffmpeg -i input.mp4 -vf "perspective=150:80:1130:120:80:980:1200:900:0:0:1280:0:0:720:1280:720:sense=0" -c:a copy output.mp4

Parameter Explanations for the Command

-i input.mp4: Specifies the source video file.
-vf "perspective=...": Applies the video filter with your designated coordinate mapping.
-c:a copy: Copies the audio stream directly without re-encoding to save time and preserve quality.
output.mp4: The corrected output video file.