psnr and ssim to calculate these metrics accurately. By processing both streams simultaneously, you can generate precise numerical data that reflects visual quality degradation during encoding.
Understanding Compression Metrics
When compressing video, codecs like H.264 or H.265 discard information to reduce file size. This process introduces artifacts such as blocking, blurring, or ringing. To evaluate the effectiveness of your encoding settings, you need objective metrics rather than subjective human observation. The two most common standards are PSNR (Peak Signal-to-Noise Ratio) and SSIM (Structural Similarity Index).
PSNR measures the ratio between the maximum possible power of a signal and the power of corrupting noise. Higher values generally indicate better quality, though it does not always correlate perfectly with human perception. SSIM is more advanced because it accounts for changes in structural information, luminance, and contrast. It produces a value between -1 and 1, where 1 indicates identical images. For professional workflows, combining both metrics provides a robust view of compression impact.
Setting Up the Comparison Pipeline
To measure these artifacts, you must feed both the original uncompressed video and the compressed output into FFmpeg simultaneously. This allows the filter graph to compare pixels frame by frame. You cannot use standard encoding commands alone; you need the -lavfi option or -filter_complex to handle multi-input processing.
First, ensure you have a lossless reference file. If your source is already compressed, the measurement will be flawed. A good practice is to keep a high-bitrate or uncompressed master copy. If you need to prepare a lossless intermediate for comparison, refer to our guide on FFMPEG Lossless: A Guide to Lossless Audio and Video Compression. This ensures that any differences detected are solely due to the final compression step you are testing.
Calculating PSNR with FFmpeg
The simplest metric to implement is PSNR. It requires two input files: the original and the encoded video. Use the following command structure to generate per-frame PSNR values.
ffmpeg -i original.mp4 -i encoded.mp4 -lavfi "[0:v][1:v]psnr" -f null -This command maps the video stream from the first input ([0:v]) and the second input ([1:v]) into the psnr filter. The output is sent to null because we only want the logs, not a new video file. The terminal will display lines containing the average PSNR, minimum, and maximum values for each frame. Look for the line starting with Average at the end of the log to get the overall score for the entire clip.
Calculating SSIM for Structural Analysis
For a more perceptually accurate result, use the SSIM filter. The syntax is nearly identical to PSNR but uses a different filter name.
ffmpeg -i original.mp4 -i encoded.mp4 -lavfi "[0:v][1:v]ssim" -f null -SSIM outputs a value closer to 1.0 for high-quality matches. While PSNR might suggest acceptable quality even when visible artifacts exist, SSIM often drops significantly when structural details are lost. This makes it particularly useful for evaluating text overlays, sharp edges, and fine textures. If you are converting specific camera formats, such as converting Fuji .mov Files to MP4 Using FFmpeg, using SSIM helps ensure that the color science and detail retention remain intact despite container changes.
Integrating Quality Checks into Workflows
In automated encoding pipelines, you can parse the output of these commands to reject encodings that fall below a certain threshold. For example, if your target audience watches on mobile devices, you might set a minimum PSNR of 40 dB. If the encoded video scores lower, the pipeline can flag it for re-encoding with higher bitrate settings.
You can also embed these checks alongside other operations. If you are adding audio tracks, such as adding random music for multiple videos, ensure that the video quality check runs independently of the audio muxing process. This isolates video-specific compression issues from audio synchronization problems. Similarly, when dealing with orientation issues, remember that autorotation affects pixel dimensions. Always compare frames after rotation has been applied to both source and destination to avoid false artifact detection due to resolution mismatches.
Troubleshooting Common Errors
If you encounter errors related to pixel format mismatches, FFmpeg may refuse to process the streams together. Ensure both inputs share the same pixel format, typically yuv420p. You can force this using the format=yuv420p filter before the PSNR or SSIM node. Another common issue is resolution mismatch. Both videos must have identical width and height. If they differ, use the scale filter to resize one stream to match the other before comparison.
Sometimes, users try to encode directly within the filter complex without proper output handling. Remember that libx264 is an encoder, not a filter. It cannot sit inside -lavfi or -filter_complex as a standalone node without a valid output context. Instead, perform the encoding in a separate pass or use the -c:v libx264 flag in the main command line, then compare the resulting file against the source.
Frequently Asked Questions
What is a good PSNR score for video?
A PSNR above 40 dB is generally considered excellent quality with barely noticeable artifacts. Scores between 30 and 40 dB indicate acceptable quality for most streaming purposes. Below 30 dB, artifacts become visually distracting and obvious to the viewer.
Can I use SSIM for audio quality measurement?
No, SSIM is designed specifically for image and video signals. It analyzes spatial relationships between pixels. For audio, you should use metrics like PESQ (Perceptual Evaluation of Speech Quality) or PSMR, which are tailored for frequency domain analysis.
Why do my PSNR values fluctuate between frames?
Dynamic scenes with fast motion or complex textures naturally produce varying PSNR values. Static backgrounds yield consistent scores, while action sequences cause spikes in error rates. Always look at the average value across the entire clip rather than individual frame peaks.
How do I handle different resolutions during comparison?
Both inputs must match exactly in resolution and pixel format. Use the scale filter to resize the smaller stream to match the larger one. For example, add [1:v]scale=1920:1080[ref] before feeding it into the PSNR filter.
Does FFmpeg support real-time artifact monitoring?
FFmpeg processes files sequentially, so it is not real-time in the sense of live broadcast monitoring. However, you can pipe live streams through these filters to get instant feedback on encoding parameters, allowing for rapid iteration during development.