Submission Date: November 2003

Interlacing causes complications in many image processing tasks. Theoretically, interlacing can be perfectly undone, but there remain a few difficulties in de-interlacing. Aliasing is an inherent artifact due to the interlaced sampling of video data. Interlacing also causes the loss of vertical resolution in the video. Motion in the video scenes need to be modelled accurately for the removal of aliasing. However, motion is difficult to estimate in the presence of high frequency artefacts, such as aliasing and noise. In this thesis, the affine tranformation is proposed to model global motion in the video scenes such as camera pan and zoom. A two-stage motion estimation technique is employed: block-matching and best fit of affine parameters. Weight function is used as a confidence measure for each block in the video fields to improve motion estimation by 50%. Motion estimation is needed for both the proposed de-interlacing methods: the averaging method and the matrix method. The averaging method is able to reduce noise and aliasing artefact by averaging a sufficiently large number of warped fields corresponding to a particular scene. Edge-dependent interpolation may be used in the averaging method to increase the vertical resolution lost in the interlaced video. On the other hand, the matrix method uses Wiener solution to obtain a best estimate of the reconstructed frame based on an appropriate choice of statistical models for the image and noise. The matrix method achieves a 30% improvement as compared to the averaging method.