Method for recognizing objects in an image pixel plane

Claim 1

A method for multiple object recognition on an image pixel plane of received images, said method comprising: (a) roughly classifying all pixel points of said received images according to whether or not a pixel point is relevant for said multiple object recognition, according to a set of first criterion related to a nonhomogeneous vicinity around each particular pixel point, to eliminate irrelevant pixel points from the relevant pixel points; (b) forming a reduced image based on said relevant pixel points as roughly classified in step (a); (c) filtering each reduced image for forming at least two filtered images whereby image components relevant for said multiple object recognition are retained in said at least two filtered images; (d) further classifying each pixel point of said at least two filtered images for providing classified images, wherein said further classifying is performed by a group of different classifiers which operate in accordance with learned rules to allocate, with evaluation number, said classified images to different object classes, wherein each classifier of said group of different classifiers operates based on a characterizing vector of a respective filtered image forming a direct input information for its respective classifier and wherein each different classifier works independently of any other classifier of said group of different classifiers; (e) merging, for each pixel point, said classified images in accordance with an algorithm based on the evaluation numbers to perform a combined global evaluation for each class of said different object classes for providing merged images; and (f) deciding for each pixel point, on the basis of said merged images, whether a pixel point of all remaining pixel points is still relevant and if so to which of said different object classes each relevant pixel point belongs.

Claim 2

A method for multiple object recognition on an image pixel plane of received images, said method comprising: (a) roughly classifying all pixel points of said received images according to whether or not a pixel point is relevant for said multiple object recognition, according to a set of first criterion related to a nonhomogeneous vicinity around each particular pixel point, to eliminate irrelevant pixel points from the relevant pixel points; (b) forming a reduced image based on said relevant pixel points as roughly classified in step (a); (c) filtering each reduced image for forming at least two filtered images whereby image components relevant for said multiple object recognition are retained in said at least two filtered images; (d) further classifying each pixel point of said at least two filtered images for providing classified images, wherein said further classifying is performed by a group of different classifiers which operate in accordance with learned rules to allocate, with evaluation number, said classified images to different object classes, wherein each classifier of said group of different classifiers operates based on a characterizing vector of a respective filtered image forming a direct input information for its respective classifier and wherein each different classifier works independently of any other classifier of said group of different classifiers; (e) merging, for each pixel point, said classified images in accordance with an algorithm based on the evaluation numbers to perform a combined global evaluation for each class of said different object classes for providing merged images; and (f) deciding for each pixel point, on the basis of said merged images, whether a pixel point of all remaining pixel points is still relevant and if so to which of said different object classes each relevant pixel point belongs.