Entropy of Markovian processes: Application to image entropy in computer vision (Q5959128)

For image processing, the author introduces about twelve new measures of information, for example the density of conditional entropy of \(f\) as \(\frac{1}{2} \int_0^{2\pi} \ln (f'_x\cos\alpha +f'_y\sin\alpha)^2 d\alpha.\) Two propositions and five lemmata are offered (the lemmata do not lead to the propositions). The style is conversational, with (triple) exclamation marks; the statement of Lemma 2 is ``As a direct consequence of the definition of discrete Markov processes, by using entropy with respect to a measure, one can obtain the entropy of the trajectory \(x(t)\) on the finite interval \([0,T],\) in the form \(H(x(\;);0,T)=H(x_0)+\int_0^T\int_{\mathbb{R}^n} p(x,t)\ln[(2\pi e)^n|GQG^T|]^{1/2} dx dt\), where \(p(x,t)\) is the probability density defined by the Fokker-Planck-Kolmogorov equation \(\frac{\partial p}{\partial t}= -\sum_{i=1}^n\frac{\partial(pf_i)}{\partial x_i}+\frac{1}{2}\sum_{i,j=1}^n \frac{\partial^2[p(GQG^T)_{ij}]}{\partial x_i\partial x_j}\)''.