TIFF different than BMP in image compression

hello every one ,

i use lena.tiff color image and gray image and i compressed it.

what is the different if i use lena.bmp, which one is better quality and better compression ? is there any different?

thanks for all

Answer 1

[DGM]

I'm assuming you're asking about picking a format when writing an image file. The answer depends, but generally, TIFF would be the better choice.

Both are potentially lossless, so they're both capable of the same quality. TIFF can utilize different compression depending on the options, some of which are lossy. The default should be lossless, though.

As far as which offers better compression, that'd be TIFF, since imwrite can only write uncompressed BMP.

TIFF also offers a lot of other capabilities which BMP does not. That may or may not matter for your usage.

That said, PNG is also a lossless compressed format and is widely supported.

If you're unsure which to use, or which settings to use, you can always write some sample images. Observe the relative file sizes, and read them back to verify that the conversion was indeed lossless.

[eng m]

thanks DGM for your answer

[Image Analyst]

The default with TIFF is 'packbits' compression (not sure what the detail of that are) but it can give a size more than the uncompressed size:

rgbImage = imread('peppers.png'); % Read in standard color demo image.
imwrite(rgbImage, '0001peppers.tif'); % 580 KB using default 'packbits'
imwrite(rgbImage, '0002peppers.tif', 'Compression', 'none'); % 577 KB
imwrite(rgbImage, '0003peppers.png'); % 281 KB

So this is one reason why I prefer PNG. Many, many other people do too. It seems to have become a defacto standard. Seems like everyone is using it now. File size is half of what TIFF provides as you can see from the sizes i gave above.

[eng m]

thank you image analyst, i read image in "lena.TIFF" then after compressed with DCT transform i write it lena.jpeg because the DCT deal with JPEG

if i use lena.bmp. and then i write it in jpeg, which give me small size after compression or its same.

[DGM]

I suppose Image Analyst is right. Not all compression schemes work well with all types of image content. Packbits is RLE, so it would stand to reason that it performs poorly with some photographic content. For images with lots of flat regions, it does better.

% try a photographic image
A = imread('cameraman.tif');

imwrite(A,'bmptest.bmp')
S = imfinfo('bmptest.bmp'); S.FileSize

ans = 66614

imwrite(A,'tiftest.tif','compression','none')
S = imfinfo('tiftest.tif'); S.FileSize

ans = 65782

imwrite(A,'tiftest.tif','compression','packbits') % default
S = imfinfo('tiftest.tif'); S.FileSize

ans = 65126

imwrite(A,'tiftest.tif','compression','lzw')
S = imfinfo('tiftest.tif'); S.FileSize

ans = 54984

imwrite(A,'tiftest.tif','compression','deflate')
S = imfinfo('tiftest.tif'); S.FileSize

ans = 48062

imwrite(A,'pngtest.png')
S = imfinfo('pngtest.png'); S.FileSize

ans = 38267

Compare the relative file sizes to those when the image has contiguous flat regions.

% what if the image is quantized?
A = (A/16)*16;

imwrite(A,'bmptest.bmp')
S = imfinfo('bmptest.bmp'); S.FileSize

ans = 66614

imwrite(A,'tiftest.tif','compression','none')
S = imfinfo('tiftest.tif'); S.FileSize

ans = 65782

imwrite(A,'tiftest.tif','compression','packbits') % default
S = imfinfo('tiftest.tif'); S.FileSize

ans = 30790

imwrite(A,'tiftest.tif','compression','lzw')
S = imfinfo('tiftest.tif'); S.FileSize

ans = 14882

imwrite(A,'tiftest.tif','compression','deflate')
S = imfinfo('tiftest.tif'); S.FileSize

ans = 13732

imwrite(A,'pngtest.png')
S = imfinfo('pngtest.png'); S.FileSize

ans = 15609

But yeah, PNG is generally better right out of the box.

[DGM]

If you're reading in two different images and then writing them again, then I can't generally make any claim which will be smaller. From my perspective, I cannot know if the two images you have are the same. If they are the same, then the source format is irrelevant. If they're not the same, then the result will differ depending on how the sources differ and to what degree.

[eng m]

thanks very much

Answer 2

[Walter Roberson]

i read image in "lena.TIFF" then after compressed with DCT transform i write it lena.jpeg because the DCT deal with JPEG

No, that is not correct reasoning.

When you use imwrite() and ask for jpeg and do not ask for Lossless, then whatever data you ask to write will be put through jpeg lossy compression. Unless you used the exact sample computation as imwrite() does, including details about the rounding strategies (an important but often overlooked detail for jpeg), then whatever you carefully arranged in your array will be changed as it is written as jpeg.

Using imwrite() to a jpeg file is not the right way to do DCT compression and write out the calculated coefficients as a JPEG file!! You cannot do, for example,

im = imread('flamingos.jpg');
for K = 1 : 3
  B{K} = dct2(im(:,:,K));
end
BB = uint8(cat(3,B{:}));
imwrite(BB, 'dct.jpg');

and expect the result to be the JPEG version of the original image!

In order to use imwrite(), you have to have reconstructed the image, like

im = imread('flamingos.jpg');
for K = 1 : 3
  B{K} = do_my_dct_compression(im(:,:,K));
end
for K = 1 : 3
  C{K} = do_my_dct_reconstruction(B{K});
end
CC = uint8(cat(3,C{:}));
imwrite(CC, 'dct_reconstruction.jpg');

but at that point, CC is the reconstructed image, which "ideally" would be identical to the original image, but instead has probably discarded some details.

imwrite() has no knowledge that CC has already gone through DCT deconstruction and reconstruction, and will treat CC just like any other array, and will process it through its own compression routines.

The only time you can "win" with doing your own compression and reconstruction is if your compression + reconstruction code discards details in a way that is compatible with the compression that will be done by imwrite, except that your compression has to have been "better" at it. For example, if you discarded more DCT components than imwrite() would do, and you happen to align your blocks exactly the same way that imwrite() would, then imwrite() might find more discardable zeros than it would have found for the original image unchanged. It would not be the case that you were writing the compression components that you had calculated: imwrite() would still write the compression components that it calculated, but you would have carefully calculated those in a way that would "prime" imwrite() to be able to do a better job than it would normally do on the original image.

If you want to write out your own compression components, if you are trying to compress the image "yourself", then you need to know the details of the image format, and you would need to write a complete binary file that happened to match the standards for the kind of image you are trying to imitate. Sometimes though, you can find third-party libraries similar to libjpeg that are willing to accept raw DCT components and write them out in a proper container -- but in such a case you have to be sure that the raw DCT components you are sending are compatible with the ones that would normally be written.

[eng m]

thanks for explain

i read the image in TIFF but in reconstructed image i write it in jpeg (the original image and reconstructed ) to know compressd size.

i want to know if i read the image in anthor format what happened in size ,which one better

[Walter Roberson]

That doesn't work, not at all.

Suppose you had a world-champion suitcase packer, able to pack a drawer of clothes into a smaller suitcase than anyone else can. Except, for competition purposes, "suitcase" is a cardboard box rather than a commercial suitcase with handles and locks and wheels and so on -- more the "idea" of a suitcase than something you can take on an airplane.

So, you watch as the champion suitcase packer carefully puts the clothes into the very small suitcase, using vacuum bagging as part of the technique to get clothes very small. And then when you want to measure how well the champion did, imagine that instead of measuring the size of the competition-style suitcase, you had the champion take all the clothes out of competition suitcase and out of the vacuum bags, and put the clothes into a commercial suitcase, except this time without being allowed to use vacuum bags. And then you measure the size of the commercial suitcase (the one where vacuum bags were not permitted), and say that that is how well the champion packer did on their competition suitcase when they used vacuum bags.

You are measuring two different things. If you want to know how well your DCT compression did in memory, then you need to look at amount of memory that would be required to store everything needed to reconstruct from your DCT compression. You would not expand your DCT compressed version back to an array and then ask how well JPEG can compress the reconstructed version.