I am working with Wavelet + SPIHT based image compression.For cameraman image there are getting different PSNR values for different images.
What about the compression ratio.Will there be any change in compression ratio for the same image while using different wavelets.?
As far as I know the main idea of wavelet-based compression is that after wavelet transform image content is redistributed in such a way that information about its regions with low spatial frequency (e. g. quasi-homogeneous regions) is concentrated in low frequency wavelet coefficients, while information about textures and noise present in the processed image is concentrated in high-frequency wavelet components. Wavelet components are compared with some threshold established taking into account noise parameters.All the components with values less than the threshold are set to zero. When you use different wavelet bases you may obtain a littele different values of coefficients, so number of coefficients set to zero may vary. That results in different compression ratio.
Same as above and I agree, same image with different wavelet always have significant different...
In fact the wavelet decomposition coefficients are obtained by correlating the image with the 2D wavelets which act as a set of base functions. This means that the absolute value of the wavelet coefficients depend on how the image correlates with the wavelet at different scales. These are the values to be compared to the thresholds as part of the compression process (as it is explained in the previous answer) and therefore there is a dependence of the compression ratio upon the specific wavelet used. This means also that probably you could find a "best" wavelet for the compression of a specific class of images.
Usually, you select the best wavelet from two points of view: Energy compaction and whether you want lossless coding or not. In the first case you select the filter bank CDF 9/7 and in the second case you use a lossless filter bank 5/3 (see jpeg2000). But if you are not interested lossless go for CDF 9/7, it is the best FB from the energy compaction point of view and toy will obtain excelent coding gain:
http://ieeexplore.ieee.org/stamp/stamp.jsp?arnumber=7081541
http://electronicimaging.spiedigitallibrary.org/article.aspx?articleid=2293271
@Juan V. Lorenzo-Ginori : Sir, How much will be the differences in compression ratio for different wavelets, if we use the simple DWT+Spiht based GRAYSCALE image compression.
I just simulated it and obtained different compression ratios while using different wavelets. But the difference means it is only in decimal points
@Victoriya V Abramova : Madam, I just simulated it and obtained different compression ratios while using different wavelets. But the difference means it is only in decimal points.
The gain depends on the image, usually you use test images and compare them using the PSNR and SSIM for diffferent bitrates (compression ratios). There are some papers that compare this for different images and there you can take them as a reference.
@Humberto Ochoa : In a compression,if the input parameter bpp = 1, will there be any compression ratio
yes, in average you are getting 1 bit per pixel, if your original image is 8 bits per pixel you are getting a CR of 8 to 1
Look, usually your CR is the uncompressed file size of your image divided by the compressed size. For example, if your raw uncompressed image size (no header) is 100 MB and your compressed file is 10 MB you have a CR of 100/10 = 10. You have an explicit CR of 10:1.
In general, the CR is expected to be the same under different wavelet filters. However, it could vary but very slightly when changing the filter. I recommend to double check you code first.
@Hazem Munawer Al-Otum : I am getting the CR with very little difference
- Badges
- Science topic
- Similar topics
- Computer Science
- Images