What is the typical Compression ratio for color images?

Sir, Iam working on fft algorithms for color images.I found that the number of bytes in fft file is more than no.of bytes in the original image and hence compression ratio is less than 1. Iam not able to make out whats my mistake. I have found the fft of image and then finding the size of the fft file. If I do quantization the compression ratio is 1.3 but I heard it should be 30-40. I have written an fft functio I am able to reconstruct the image and psnr is above 35dB with the designed fft function.

I want to know how to find compression for color image. Plz help me

[Walter Roberson]

What is an "fft file" ? What theoretical reason is there to expect that the fft file might ever be smaller than the original image size?

Answer 1

[Anitha Raj]

Sir, Firstly I thank for the response. For fft,I meant fast fourier transform. I have designed a new algorithm for Fast Fourier Transform and applied on color images.The PSNR is above 35dB , the original & reconstructed images are indistinguishable.But when I quantize(using quantization matrix)) & encode(filtering all zero coeffients) for compression the compression ratio is 1-1.5. CR=(Image size/length of non zero co efficients in fft domain).What is the typical compression ratio(ie.,original image to its transform image) for color images?

[Image Analyst]

Why don't you just take some color images and save them in various compressed formats (PNG, compressed TIF, JPG, JPEG2000) and see for yourself. Like I said, I usually see that PNG are about a third as big. It really depends on how much variation there is in your images. You may get great compression for slowly varying, very smooth images and not so great for images with a lot of high frequency details. A uniform image could compress down to three values (number of rows, number of columns, and the value), or even one value if you make some assumptions about the number of rows and columns.

[Walter Roberson]

If you are quantizing then you are using lossy compression, not lossless compression. Questions about average lossy compression are quite different than questions about average lossless compression.

When you say that the original and reconstructed images are indistinguishable, which measures are you using for that? For example how is your algorithm at preserving the data of astronomical images?

[Image Analyst]

Anitha's "Answer" moved here since it's a comment, not an answer.

Sir, Iam finding the PSNR of images and it is above 35dB.There are no artifacts in the reconstructed images. I ve not tested on astronomical images. My algorithm is using lossy compression method,still I am getting very less compression.What may be the mistake.Please help me

[Walter Roberson]

When you are saying that there are no artifacts in the reconstructed images, do you mean that isequal() of the two images will be exactly the same?

Answer 2

[Walter Roberson]

The typical compression ratio for color images is 512:514, just a little below 1.

The theoretic typical compression ratio for color images is arguably 1:1 exactly, but is also arguably only that asymptopically, with all files (compressed or uncompressed) arguably required to add one bit which signals whether the file is a compressed image or an uncompressed image. Yes, that one extra bit really does make a difference in the field of compression theory -- makes a big difference.

Answer 3

[Image Analyst]

Of course an fft image will be more bytes than a spatial domain image - that's obvious because there are 8 or 16 bytes per pixel instead of 1. The fft gives complex double precision values.

I find for my images that a PNG color image is about 1/3 the size of a TIF or BMP version. JPEG versions are even smaller but of course they look crummy.

[Walter Roberson]

Right, those values are for your images, which are images of interest to you. But over the set of all color images, the theoretical rate approximates 1:1.