An iterative elastic stretching technique applied to thermographic images

A new approach has been proposed for the automatic matching of two or more objects based on gradual and sequential deformation of an elastic image onto a target image. It is based on the belief that cooperative and gradual matching minimises mismatches due to ambiguous relationships between features. An efficient but simple nearest neighbour rule is used to supply sufficient information to correctly match simple objects such as hands. However, with very busy and/or grossly deformed and irregular images more expensive feature measures may be necessary. These could be based on gray-scale intensity, colour, texture or orientation. Though this automatic method does not use information about local shape, orientation and intensity it succeeded in matching efficiently and successfully irregularities caused by movements of the object. An enhanced frequency domain version of the technique has also been developed. Although fast Fourier transformation techniques are widely used in signal processing and image restoration, to date they have not been applied in matching techniques as proposed in this paper