Transcoding - definition
Transcoding is the process of decoding a digital image or video file and encoding it again, usually with a different codec, bitrate, resolution, frame rate, color format, or delivery container. It is different from remuxing, which changes the container format without decoding and re-encoding the media stream.
For image and video anonymization, transcoding matters because the output encoding can change the visual appearance of blurred faces and license plates. A blur or mask should be applied to decoded image pixels before the final output is encoded. The final anonymized file must then be reviewed in the same form in which it will be shared, stored, or published.
Common video codecs used in transcoding include Advanced Video Coding (AVC), specified in ITU-T Recommendation H.264, and High Efficiency Video Coding (HEVC), specified in ITU-T Recommendation H.265. These codecs use prediction, transform coding, quantization, and entropy coding. Those processes reduce file size, but they can also create block edges, ringing, color changes, and temporal artifacts around anonymized regions.
Why transcoding affects face and license plate blurring
Blurred areas are not separate privacy controls once they have been encoded into the output video or image. They become part of the pixel data. A later transcode can change their appearance, particularly at the boundary between the blurred area and the unmasked background.
A transcoder does not recover original pixels that were removed by an effective blur or opaque mask. However, it can expose weaknesses in an anonymization workflow. For example, a blur that appears sufficient in a high-bitrate master file may become uneven after aggressive bitrate reduction, resizing, sharpening, or repeated encoding.
Transcoding event | Potential effect on anonymized footage | Practical control
|
|---|---|---|
Lower bitrate encoding | Block artifacts or ringing may make blur boundaries more visible or make the footage difficult to assess. | Review the final delivery file at normal and enlarged viewing sizes. |
Resolution reduction | A face or license plate may occupy fewer pixels, while a mask may also become too small or misaligned at the edge. | Apply masking before resizing and verify coverage after resizing. |
Resolution enlargement | Interpolation can soften or reshape mask edges without restoring removed source pixels. | Use the final intended resolution for quality assurance. |
Repeated lossy encoding | Each generation can add artifacts and reduce overall image readability. | Keep one protected source and create delivery versions from a single anonymized master. |
Frame-rate conversion | Frame interpolation or dropped frames can affect tracking continuity in moving footage. | Inspect mask coverage during motion, cuts, and scene changes. |
Transcoding workflow for anonymized images and video
A privacy-preserving workflow separates source handling, masking, encoding, and verification. The order is important because encoding settings should not be treated as a substitute for face blurring or license plate blurring.
- Decode the source image sequence or video into frames.
- Detect faces and license plates in the decoded footage.
- Track detected areas across frames and correct missed or inaccurate detections.
- Apply a blur, pixelation effect, or opaque mask with sufficient coverage around the target area.
- Encode the anonymized result once for the intended delivery format.
- Review the encoded output, including fast movement, low-light scenes, camera cuts, and compression-heavy segments.
- Store the original and the anonymized derivative under separate access controls where the original must be retained.
Gallio PRO is designed for processing stored images and video files. It automatically detects faces and license plates, while other visual elements, such as documents, name badges, tattoos, logos, or monitor content, require manual masking in the built-in editor. A final transcode should therefore be assessed for both automatically and manually masked regions.
Key transcoding parameters and metrics
Codec settings affect readability, storage requirements, and the visibility of anonymization artifacts. A quality metric can help compare encodes, but it cannot by itself prove that a face or license plate is adequately anonymized.
Parameter or metric | Meaning | Relevance to anonymization
|
|---|---|---|
Codec | The compression method used to encode the media, such as AVC or HEVC. | Different codecs produce different artifact patterns at similar file sizes. |
Bitrate | The amount of encoded data per unit of time. | Lower bitrates generally require stronger compression and can reduce image readability. |
Quantization parameter | An encoder control associated with quantization strength. | Its scale and interpretation depend on the codec and encoder implementation. |
Chroma subsampling | The reduction of color detail relative to brightness detail. | It can alter colored blur boundaries and license plate regions. |
Peak Signal-to-Noise Ratio (PSNR) | A pixel-error-based comparison between a reference and encoded image. | Useful for encode comparison, but not a measurement of privacy protection. |
Structural Similarity Index Measure (SSIM) | A metric designed to assess structural similarity between images. | It can indicate visual change, but cannot determine whether a person is identifiable. |
PSNR is commonly expressed as PSNR = 10 × log10(MAXI2 / MSE), where MAXI is the maximum pixel value and MSE is mean squared error. The metric measures encoding difference, not whether the masked content remains protected. The limitations of pixel-based quality metrics are discussed in the original SSIM publication by Wang et al. (2004).
Challenges and limitations of transcoding
The main technical risk is not that a standard codec reverses a properly applied mask. The risk is that the masking method, tracking, or final output review is inadequate for the footage and the intended disclosure context.
- A small or weak blur may leave visual detail that was never removed from the source frame.
- Motion blur, occlusion, reflections, and scene cuts can cause a detected face or license plate to be missed in one or more frames.
- Encoding artifacts can make manual review harder, especially in dark scenes or fast-moving footage.
- A file uploaded to a platform may be transcoded again by that platform, creating a further derivative that should be evaluated where possible.
- Sharpening, denoising, and artificial intelligence-based enhancement applied after anonymization can change the appearance of masked regions and should not be assumed to preserve the original privacy assessment.
Standards and references
Transcoding is implemented through codec specifications and encoder behavior. The following sources define or explain the technical components most relevant to anonymized video output.
- ITU-T Recommendation H.264 - Advanced video coding for generic audiovisual services, International Telecommunication Union, 2021.
- ITU-T Recommendation H.265 - High efficiency video coding, International Telecommunication Union, 2021.
- ISO/IEC 23008-2:2020, Information technology - High efficiency coding and media delivery in heterogeneous environments - Part 2: High Efficiency Video Coding.
- AV1 Bitstream and Decoding Process Specification, Alliance for Open Media.
- Image quality assessment: From error visibility to structural similarity, Wang, Bovik, Sheikh and Simoncelli, IEEE Transactions on Image Processing, 2004.