save fonction and compression

Hello!

I see something interesting about the 'save' function: I tried to code a huffman code in order to create a JPEG compression, I finish all technics (DCT, quatization, zigzaig, MLEM, etc...), and i'm stuck about Huffman's Entropy coding.

But I realyse sompething when I save the image's data with "save" fuction: Thie size of the compressed image (starting .bmp 6.000Ko) is reduce as a size like the same than the jpg version (600Ko), so: save function in .m file use the Huffman technique? How can I see the codesource of "save" function?

Answer 1

[Baptiste]

I talked about save() function . Ok, obviously, Matwork don't use a common compression algorithm and his function is too particular.

But If I success the Humann coding, and I must save those data, instead of save() who compress a already compressed files, I must use imwrite() or saveas()?

Anyway, I'll try to understand what this Huffman wanted to do with those AC & DC's RLE data for JPEG, I think I will don't sleep during 48 hours -_- .

Thank you for your answer!

[Walter Roberson]

No, Mathworks uses a very common compression algorithm. According to their documentation for the older-style .mat files (not -v7.3),

To decompress the contents of a compressed variable in a MAT-file, you can use the uncompress function from the freeware zlib-1.1.4 library available at the gzip web site,  http://www.gzip.org/zlib/ .

According to wikipedia,

gzip is based on the DEFLATE algorithm, which is a combination of LZ77 and Huffman coding.

so I was wrong about Huffman encoding in part, but it is important to note that the LZ77 algorithms from the Lempel–Ziv–Welch family are strong compressors beyond what Huffman itself could give.

[Walter Roberson]

To avoid re-compressing an already compressed binary stream, use fwrite() to write the binary out to a file.

When you are writing a Huffman type compression, remember to take care of the problem that files must store a whole number of bytes but the last Huffman encoding might end in the middle of a byte.

[Baptiste]

I wasn't looking the wikipedia page, now it's clear!

Thank you very much!

Answer 2

[Walter Roberson]

The source of "save" is Mathworks proprietary, and it would be very expensive to get access to it unless you get a job with Mathworks.

I can pretty much guarantee that they do not use Huffman encoding. If save() is compressing data at all, it is likely an Lempel–Ziv–Welch type algorithm.

But please check to see if you are talking about save() or about saveas() or about imwrite() as save() does not create .bmp files but the other two do.