A priori probabilities in image classification for the estimation of surface areas by the regression methodFunded under: JRC-REMSENS 4C
High resolution satellite images (SPOT-XS and LANDSAT-TM) are increasingly used to estimate surface areas (in agricultural statistics, environmental studies etc.). The controlled classification of images is normally done using discriminant analysis by maximum likelihood on the assumption of normality, although alternative methods, such as the convex envelopes method, have been tested. Maximum likelihood with equal prior probabilities frequently underestimates larger classes and overestimates small ones. Prior probabilities proportional to the frequencies have the opposite effect and can even lead to the disappearance of some classes.
Bibliographic Reference: Paper presented: XXIII Journées de Statistique, Strasbourg (FR), May 27-30, 1991
Availability: Available from (1) as Paper FR 35974 ORA
Record Number: 199210389 / Last updated on: 1994-12-02
Original language: fr
Available languages: fr