Noggranna bildbaserade mätningar genom oregelbunden sampling

Tidsperiod: 2015-01-01 till 2018-12-31

Projektledare: Gunilla Borgefors

Finansiär: Vetenskapsrådet

Bidragstyp: Projektbidrag

Budget: 3 643 000 SEK

Mathematical morphology (MM) is a widely-used framework for efficient processing and analysis of images. Linear filters, according to the Nyquist-Shannon sampling theory, are not affected by the sampling grid. This is not the case for MM operators. That is, the precise location of an object within the camera´s field of view affects the measurement results obtained through MM. To avoid this, an alternative sampling theory would be needed. As a first step towards such theory, we propose to increase the sampling density at relevant locations within the result of the operator. This requires an alternate image representation using graphs, on which MM operators are already defined. The result will not be complete independence of the sampling grid, but a significantly reduced influence, and thus a significantly increased precision in measurements and other results.A second application for the graph-based irregular image representation is data obtained from range cameras, which are becoming increasingly common. Data from range cameras are irregularly sampled. These data are typically first resampled on a regular grid, and larger holes, caused by the surface geometry, need to be patched. By converting the irregularly sampled data to our graph representation, the holes can be maintained. The MM tools defined within this project will be directly applicable to this type of data, yielding more efficient and precise analyses.