![]() | ![]() |
Homer.ppm (1,467,663 bytes) | Homer.j2c (12,101 bytes (122 times smaller)) |
(Y’CbCr)=(0,2990,5870,144−0,1687−0,33130,50,5−0,4187−0,0813)(RGB) | (1) |
which when the dinamic range of the RGB components is [0,255] can be approximated by
(256×Y’256×(Cb−128)256×(Cr−128))=(7715029−44−67131131−110−21)(RGB) | (2) |
in order to avoid the floating point computations.
R, H(R)=7.51 bpp | G, H(G)=6.82 bpp | B, H(B)=7.04 bpp |
![]() | ![]() | ![]() |
Total = 21,37 bpp |
Y’, H(Y’)=7.42 bpp | Cb, H(Cb)=6.86 bpp | Cr, H(Cr)=4.51 bpp |
![]() | ![]() | ![]() |
Total = 18,79 bpp |
PSNR[dB]=10log10(2b−1)2MSE, | (3) |
where b is the number of bits/pixel and MSE (Mean Squared Error) is
MSE=1NN∑i=1(s[i]−ŝ[i])2, | (4) |
where N is the number of pixels, s[i] is the i-th (i=(x,y)) point of the s[⋅] image and ŝ[i] is the i-h point of the reconstructed image.
![]() | ![]() |
Original (4:4:4) | Subsampled (4:2:0), PSNR =22.07 dB |
![]() | ![]() |
Original (8:8:8) | Subsampled (8:2:0), PSNR =22.01 dB |
![]() | ![]() |
Original (16:16:16) | Subsampled (16:2:0) PSNR =21.96 dB |
DCT[u]=√2√NK(u)N−1∑n=0s[n]cos(2n+1)πu2n | (5) |
and the inverse transform is
s[n]=√2√NN−1∑u=0K(u)DCT[u]cos(2n+1)πu2n, | (6) |
where N is the number of pixels, and s[n] denotes the n-th pixel of the image s, and
- - -
- - -
- - -
- - -
- - -
Where
S=(↑2(L)∗sL)+(↑2(H)∗sH) | (7) |
and
L=↓2(S∗aL)H=↓2(S∗aH). | (8) |
(↓2(S))i=S2i | (9) |
and
(↑2(S))i={Si∕2if i if even0otherwise. | (10) |
where Si if the i-th sample of S.
Hi=S2i+1−𝒫({S2i})i | (PredictionStep) |
Li=S2i+{𝒰(H)}i | (UpdateStep) |
{S2i+1}=↓2(Z−1(S)) |
and
{S2i}=↓2(S), |
where Z−1 represents the one sample delay function.
E(x)=∑i|xi|2, | (11) |
where |xi| represents the Euclidian Norm (also known as the L2 Norm) of the sample xi, that in a general case could be a complex number.
E(SHb)=√∑i|SHbi|2. | (12) |
Li=S2i+S2i+12, | (HaarL) |
and the i-th sample of the high-frequency subband as
Hi=S2i+1−S2i. | (HaarH) |
If Lifting is used,
Li=S2i+Hi2. | (HaarLLifted) |
- - -
- - -
- - -
- - -
- - -
Li=−18S2i−2+14S2i−1+34S2i+14S2i+1−18S2i+2 | (5/3L) |
and the i-th sample of the high-frequency signal is computed by
Hi=S2i+1−S2i+S2i+22, | (5/3H) |
that, if we use Lifting, it can be also computed using less operations by
Li=S2i+Hi−1+Hi4. | (5/3LLifted) |
- - -
- - -
- - -
- - -
- - -
Hi=S2i+1−(916(S2i+S2i+2)−116(S2i−2+S2i+4)) | (13/7H) |
and (the lifted) calulus of the i-th sample of the low-frequency signal is
Li=S2i+932(Hi−1+Hi)−132(Hi−2+Hi+1). | (13) |
- - -
- - -
- - -
- - -
- - -
[1] M. D. Adams and F. Kossentini. Reversible Integer-to-Integer Wavelet Transforms for Image Compression: Performance Evaluation and Analysis. IEEE Trans. Image Process., 9(6):1010–1024, 2000.
[2] M.D. Adams. Reversible Wavelet Transform and their Application to Embedded Image Compression. PhD thesis, A,A,Sc, University of Waterloo, 1993.
[3] A. Haar. Zur Theorie der orthogolanen Funktionen-Systeme. Mathematische Annalen, 69:331–371, 1910.
[4] The Joint Photographic Experts Group (JPEG). Recommendation T.81: Digital Compression and Coding of Continuous-tone Still Images. International Telecommunication Union (ITU), September 1992.
[5] A.M. Marcos. Compresión de imágenes. Norma JPEG. Editorial Ciencia 3, 1999.
[6] Majid Rabban, Rajan L. Joshi, and Paul W. Jones. The JPEG 2000 Suite, chapter JPEG 2000 Core Coding System (Part 1). WILEY, 2009.
[7] Ana Sovic and Damir Sersic. Signal decomposition methods for reducind drawbacks of the dwt. Engineering Review, 32(2):70–77, 2012.
[8] W. Sweldens and P. Schröder. Building Your Own Wavelets at Home.