Transform Coding Distortion

Lemma (Arithmetic-Geometric Mean Inequality)

For any sequence of kk nonnegative real numbers b1,,bkb_{1},\dots,b_{k} (i=1kbi)1k1ki=1kbi\left( \prod_{i=1}^{k}b_{i} \right)^{\frac{1}{k}} \le \frac{1}{k}\sum_{i=1}^{k}b_{i}where equality holds if and only if bi=b,i=1,,kb_{i}=b, i=1,\dots,k.

Lemma (Hadamard’s Inequality)

Let R\mathbf{R} be a positive semidefinite matrix with diagonal elements rii,i=1,,kr_{ii}, i=1,\dots,k. Then detRi=1krii\det \mathbf{R}\le \prod_{i=1}^{k}r_{ii}If R\mathbf{R} is positive definite, then equality holds if and only if R\mathbf{R} is diagonal.

Definition (Transform Coding Distortion)

A transform coder with the KLT and optimal bit allocation we define the distortion to be Dtc=khg22bˉdet(RX)1/k=khg22bˉ(i=1kλi)1/kD_{\text{tc}}=kh_{g}2^{-2\bar{b}}\det (\mathbf{R_{X}})^{1/k}=kh_{g}2^{-2\bar{b}}\left( \prod_{i=1}^{k}\lambda_{i} \right)^{1/k}where λ1,,λk\lambda_{1},\dots,\lambda_{k} are the eigenvalues of RX\mathbf{R_{X}}.

Linked from