IMAGE RECOGNITION OR RECONSTRUCTION WITH PRINCIPAL COMPONENT ANALYSIS AND WITH EXCLUSIONS OF OCCLUSIONS AND NOISE

摘要

An image is processed for recognition or verification purposes (such as the image of a person's face), or in order to reconstruct the image. Processing means (5) serve to identify picture elements representing occlusions or noise and exclude them from the comparison process (10) or from the reconstruction. This is achieved by reference to stored data (8) defining for each particular picture element position an acceptable range of values, these stored data being derived from a reference population of images. The verification or reconstruction may use the method of principal component analysis by reference to stored eigenpictures. The range data can be estimated from the eigenpictures and corresponding eigenvalues rather than calculated from the original reference images.