Savchenko A. Video compression based on nonliner multiscale decomposition

The thesis is devoted to the development of new and effective methods of video compression for transmission over communication channels. The analysis of modern methods of video and photo compression is carried out in the thesis, advantages and disadvantages of these methods are shown as well. An algorithm for construction of the proposed method is performed. This method differs from the known methods which were adopted earlier in the discrete cases. It is known from theoretical sources that RMS approximation of functions does not always exist. Uniform approximation of functions is always exists, but it requires large computation efforts and is not always acceptable for practical implementation. The basic idea of the proposed method consists in nonlinear multiscale decomposition of the original image on the basis of the fractional rational approximation. A method of approximation is offered. Approximated input matrix is compressed twice, the result of it is N / 2 - approximation coefficients. In order to get the desired result of approximation it it necessary to make from 2 to 5 iterations. In order to find N-detailing coefficients, approximating nonlinear function should be subtracted from input function. Each of the following step reduces the number of samples by half both in rows and columns. Recovering of the original images from the video container is performing by step-by-step interpolation of the last approximation function with the addition of all the detail coefficients obtained during the compilation procedure. While running a modeling process a peak signal to noise ratio measure was used in order to determine a compression quality of compared methods.