In this PhD thesis, different methods to analyse video sequences are proposed,
with the aim of applying them to the restoration of old movies. This
restoration consists of removing those artifacts that, due to the material degradation along time, the bad
conservation of the movies, and mechanical errors, lead to an important
degradation of film quality. The restoration process must protect the correct
original information as much as possible. We have focused  in the
treatment of some of these artifacts, contributing with novel solutions to the problems that have arisen us. The studied subjects have
been: 

1. Segmentation in shots of degraded sequences has been taken as a
starting point of the
restoration process. We have presented several methods, based on brightness
differences between images that belong to the compressed sequence. These methods are able to detect the abrupt transitions or cuts with a high detection
probability and a low false alarm probability, even in the case of low
quality images. We also present some methods for detecting  gradual
transitions, like dissolves and wipes. 
2. Flicker correction. In order to solve this artifact, very common in black and white
movies, two methods are proposed: the first one based on an affine
intensity transformation which reduce the random and periodic
variations of mean and variance, calculating the transformation
parameters with a temporal variation model of both
statistics. The second method
tries to improve the visual results using a nonlinear
transformation based on an histogram matching. The purpose is to
change each image histogram into a target histogram calculated as
average of the image histogram and its neighbours.

3. Noise reduction inside each shot. We present a FIR filter, with a high
computational efficiency, that reduces significantly the stationary noise. This
filter is based on the spatio-temporal response of human visual system, achieving a good
reduction factor, without introducing a visible degradation. 

4. Detection and interpolation of missing data (blotches). Three methods
are presented based on morphological operators without motion estimation. The
proposed methods are used as in the detection phase as in the
interpolation one. In order to restore  this kind of
artifacts we have exploited the blotches characteristics: high temporal variation, high contrast and low spatial variance.