Jpeg - How lossy is this image format?

L

lamap

Luis Albergaria Pacheco
Local time
4:33 PM
Joined
Apr 5, 2009
Messages
18
Dear Friends,

I have been reeding that jpeg imagem format loses quality everytime it is saved.

Is this true, when jpeg file is saved with 100% image quality?

This is the format I use the most mainly because of the lower size required to store the scan files.

Best regards,

Luis Pacheco
 
JPEG Can be used with various degrees of compression. It can be used for "Lossless" compression, and typically reduces file size to 1/4th or its original size. At higher levels of compression, it "throws away" high-frequency information.

One problem with JPEG, is that most implementations only allow for 8bits per color plane, so color is limited to 24bits per pixel. So if your camera stores 12 bit ot 16 bit values, JPEG cuts it to 8. There are JPEG schemes that store more bits, but they are proprietary and would have to be licensed.

JPEG lossless compression uses Huffman code. Many years ago, I wrote my own in FORTRAN.

http://en.wikipedia.org/wiki/Huffman_coding
 
Last edited:
JPEG compression uses Huffman, or alternatively arithmetic coding, which was however patented by IBM and didn't took off. This isn't the lossy part of it however: the loss happens during DCT which removes minor spatial frequencies. The degree of "minor" there determined by quality setting, but even at 100% there is some loss. There is lossless version which does not employs DCT, but as said, the compression ratio isn't impressive. Also, far from every piece of software is able to read and/or write lossless jpeg, as it's not part of mandatory (so-called baseline) ITU-conforming implementation.

JPEG works on blocks of 8x8 pixels, and on low quality settings you clearly can see artifacts along their sides in the image. With high quality, the loss is very minor. However the big problem, color depth (as outlined by Brian) remains.

So in short, NOT RECOMMENDED for scan originals. Use TIFF or DNG.
 
If all of the DCT coefficients are saved in the JPEG, the inverse DCT does produce the original image. The Huffman code is used to store the coefficients. The Lossy compression levels throw away the high frequency coefficients.

I've done more with Fourier transforms, but the end results are the same.
 
Just FYI, by 'every time it is saved', that means every time you open it for editing in a graphics program like photoshop and save it. If you just open it to look at, it does not degrade the file you opened, and if you are in an editing program and click 'save', then edit and click 'save' again, it does not lose data on top of lost data. People get confused about that.
 
You must log in or register to reply here.
Back
Top Bottom