3D Kalman Filtering of Image Sequences
The authors are currently engaged in processing time-series of satellite imagery. Processing includes noise elimination and in this context, a 3D Kalman filtering has been developped and is presented here.
The first step is the definition of a class of two and three-parameter Markov discrete processes. The linear filtering of such stochastic processes reduces to a one-parameter vectorial Markov process recursive filtering, described by Kalman's equations. The 3D filter is then broken down into a two-dimensional spatial filter and a one-dimensional time filter. Some more stationnarity hypothesis allows a very simplified algorithm. The CPU time required is about 4 minutes on a middle range computer, for a 512 × 512 pixels picture.
The results are exposed on one-image sequence, which is assumed to verify the Markovian assumptions.
KeywordsMarkov Process Image Sequence Observation Process Markovian Assumption Noisy Sequence
Unable to display preview. Download preview PDF.
- H. Korezlioglu, 1978 Filtrage lineaire récursif des processus à 2 indices discrets, Markoviens au sens large. Rapport ENST - D -78004, Fév. 1978.Google Scholar
- S. Attasi, 1975 Modelisation et traitement des suites à2 indices. These d’Etat ,Université Paris-VI.Google Scholar
- Huang, 1975 Topics in applied physics, Vol. 42, 2D Digital Signal Processing. springer V erlag, Berlin.Google Scholar
- H. Maître, 1980 Elements of Picture Processing. Cours ENST ,1980.Google Scholar