What lossy compression chooses not to keep
Lossy compression reduces file size by discarding some information according to a model of what matters for the intended result. An image format may preserve broad colour changes while reducing fine detail. The decoded image can look similar to the original, but the missing information cannot be perfectly recovered.
Go deeper
Smaller files through controlled approximation
Lossy compression does more than represent the same data efficiently. It changes the data by preserving some features more accurately than others.
For images, a method may reduce precision in components where errors are expected to be less noticeable. Audio methods can use models of hearing to reduce information that is less perceptually important in context.
Similarity replaces exact reconstruction as the goal. The result may be useful even though it is not identical to the original.
- Higher compression can remove more detail.
- Visible or audible errors are called compression artefacts.
- Repeated decoding and lossy re-encoding can introduce additional changes.
Simply copying a compressed file does not cause this generational loss. The problem arises when its decoded content is processed through another lossy encoding step, especially with changed settings.