In this thesis we introduce the use of differential geometry techniques in the processing  of volumetric graphical objects. The use of these techniques allow us to create a  family  of geometric filters which detect high frequencies. Such filters have the characteristics of being non-linear, isotropic and invariant with respect to scale changes. We use these filters to perform segmentation and representation of volumetric objects taking the object's high frequencies areas as a starting point.
We describe an algorithm to reconstruct the original volumetric object from the 
filtering process and we show that such method yields a perfect reconstruction. T
his 
fact is very important taking into account the non-linear nature of the geometric
 filter. 
The filtering process introduced in this work has a great potentiality in applica
tions 
that process volumetric objects. We present an application which uses such filter
s 
to acquire a temporal segmentation of a video sequence. We developed an extremely
 
robust and efficient method to detect sudden changes in a video sequence (scene b
reak ).