Research On Interpolation Algorithm Of Video Signal Base On Auto Regressive Model

With the development of multimedia communications and information processing technology,better and better image quality is demanded. However, because of the limitations of communicationband (such as wireless condition), the image quality obtained is often very low and it can't meetconsumer needs. Image interpolation can improve image quality easily and quickly. So, thetechnology is widely applied to various fields of image processing, such as image zooming, imagerestoration, image deinterlacing.This thesis mainly focuses on application of the image/video interpolation methods. Firstly,recent research achievements of image interpolation algorithms are introduced briefly. Then,according to auto regressive (AR) model of image, interpolation zooming methods of video signalare studied from image/video AR model of spatial domain and spatio-tempoal domain respectively.The main contributions of the thesis are as follows:(1) An image interpolation reconstruction algorithm based on spatial AR model is proposed.The algorithm exploits the expectation maximization (EM) algorithm to estimate the AR modelparameters with initial values obtained by image inperpolation by 2-D auto regressive modeling andsoft-decision estimation (SAI) algorithm, and then estimate the interpolation values. Theexperimental results show that PSNR of the proposed algorithm is improved 0.23dB to 0.38dBcompared with SAI algorithm, and better image visual quality is preserved .(2) Two video interpolation reconstruction algorithms based on spatio-temporal AR model areproposed. One algorithm exploits spatio-temporal redundancies to estimate paramenters of the ARmodel proposed SAI algorithm, and then estimates the image interpolation values. Experimentsdemonstrate that PSNR of the improved algorithm is 0.27dB to 0.39dB higher than SAI algorithm,and subjective performance becomes better. The other algorithm proposes a new spatio-temporalAR (STAR) model, which consists of 4 pixels from the current frame and 4 pixels from the adjacentframe, and then estimates the STAR model parameters to obtain the image interpolation values.Experiments show that the improved algorithm is able to yield the interpolated frames with highervisual performance.