BitJazz
SheerVideo Fidelity
Lossless Compression

A little background here. There are two fundamentally different ways you can reduce the size of a file: perfect-fidelity compression, and approximating "compression".

Perfect-fidelity codecs, such as SheerVideo, slim down the file by packing all the information in the file more efficiently, so it takes up less space. Perfect-fidelity compression is also known as lossless compression, in the sense of "compression without information loss"; or as nondestructive compression, because it does not destroy any information. However, the term "lossless" has often been shamelessly abused by marketeers as misleading shorthand for such nonsensical concepts as "visually lossless", "virtually lossless", or "kinda lossless", meaning "we wish it were really lossless" or "if you close your eyes, you won't notice the difference".

There are two basic techniques lossless codecs can use to compress data: removing wasted space, and entropy coding.

Removing wasted space:
Some pixel formats do not use all the available code space. For example, the 'v210' pixel format packs three 10-bit pixel components into a 32-bit word, wasting 2 out of every 32 bits. By squeezing out those unused bits, a lossless compressor can compress the data by 6.25% — not much by itself, but as an independent gain, it multiplies (not just adds) the effectiveness of any further techniques used.

Entropy coding:
Some codes are more common than others. In typical English text, for example, the letter 'e' occurs much more frequently than any other letter, while 'j' is among the rarest. In inventing the Morse code, Samuel Morse took advantage of these statistics by giving 'e' the shortest possible code (.), and 'j' the longest (. _ _ _). Thus the Morse code typically compresses messages, because the savings from giving 'e' such a short code overwhelm the losses from assigning longer codes to all the rare letters, so that the length of the message as a whole is shorter than it would be if all letter codes were equally long. More generally, it can be shown that in order to maximize compression, the optimal length of each symbol should be proportional to its entropy, which is the negative logarithm of its probability of occurrence.

optimal symbol code length ∝ -log(symbol probability)

A general problem with entropy coding is that for rare pathological cases, a compressor will actually expand a file so that it takes up more space than the original, rather than compressing it. In the Morse code example, a message consisting of nothing but 'j's would take much longer to send in Morse code than in a uniform-length code.

SheerVideo both removes wasted space and uses entropy coding. However, SheerVideo safeguards against pathological cases, ensuring that even the most unusual cases do not expand the file by more than a few bits.