Automatic classification of images

Claim 1

A method for the automatic classification of the image of a ship comprising the steps of: scanning the ship to generate a two-dimensional image thereof represented by a matrix wherein the values of the elements thereof correspond to energy received from portions of the ship, the elements of the rows of the matrix corresponding to such portions spaced substantially horizontally and longitudinally of the ship and the elements of the columns of the matrix corresponding to such portions spaced substantially vertically; summing predetermined elements of a plurality of said columns to generate a first vector in which the components thereof correspond individually to the columns of said plurality; performing a discrete Fourier transform on the components of said first vector and squaring selected coefficients resulting from said transform to generate a feature vector which is the power spectrum of the first vector and which corresponds to said ship; obtaining a plurality of such feature vectors under relatively ideal scanning conditions, the feature vectors of said plurality thereof corresponding individually to images of a ship of a predetermined class at a plurality of aspect angles and at a predetermined range; deriving from said plurality of feature vectors a decision rule corresponding to said class; obtaining such a feature vector from such an image of a potential target ship; and comparing the feature vector from the image of the potential target ship with said decision rule to classify said target ship as being in said class or as being outside said class.

Claim 2

A method for the automatic classification of the image of a ship comprising the steps of: scanning the ship to generate a two-dimensional image thereof represented by a matrix wherein the values of the elements thereof correspond to energy received from portions of the ship, the elements of the rows of the matrix corresponding to such portions spaced substantially horizontally and longitudinally of the ship and the elements of the columns of the matrix corresponding to such portions spaced substantially vertically; summing predetermined elements of a plurality of said columns to generate a first vector in which the components thereof correspond individually to the columns of said plurality; performing a discrete Fourier transform on the components of said first vector and squaring selected coefficients resulting from said transform to generate a feature vector which is the power spectrum of the first vector and which corresponds to said ship; obtaining a plurality of such feature vectors under relatively ideal scanning conditions, the feature vectors of said plurality thereof corresponding individually to images of a ship of a predetermined class at a plurality of aspect angles and at a predetermined range; deriving from said plurality of feature vectors a decision rule corresponding to said class; obtaining such a feature vector from such an image of a potential target ship; and comparing the feature vector from the image of the potential target ship with said decision rule to classify said target ship as being in said class or as being outside said class.