CODEWITHSUNDEEP
algorithmimage-processingjpeg
quantum231
quantum231

Reputation: 2585

How are code lengths limited to 16 bits at maximum in JPEG?

According to ITU T.81 that describes the JPEG format, BITS stores the "code length counts". Creation of it is described in Annex K Figure K.2 of the specification. JPEG specification expects that symbols will exist that will require huffman codes upto 32 bits in length when encoding is being carrying out. However, it limits huffman code length to 16 bits at maximum for when data is encoded. For this purpose the code lengths must be limited to 16 bits. The procedure for this is contained in Annex K Figure K.3 shown below: enter image description here

My question is that will BITS have negative values as well when we do BITS(I)-2 and BITS(I)-1? Does it have to be declared as signed? If so, what meaning do negative values have? I have implemented this in code but it gives me negative values. So some images encode just fine but others where BITS has to be manipulated to 16 bits, the images always get corrupted.

Upvotes: 1

Views: 163

Answers (1)

Nicolás Abram
Nicolás Abram

Reputation: 333

As I understand it, negative values should be fine since those are i=[17,32] values, which are not used once you are done reducing it to 16 bits. The algorithm assumes signed math, notice the BITS(i) > 0 condition, negative values will fall through the "No" branch and eventually end after dealing with BITS(17).

In your implementation, I think you could use unsigned math if you really want to and just clamp the underflow to 0 (Naively, something like BITS(i) = BITS(i) > 2 ? BITS(i) - 2 : 0).

Upvotes: 1

Related Questions