Enhanced and Parameterless Locality Preserving Projections for Face Recognition
Fadi Dornaika, A. Assoum, Abdelmalik Moujahid · Ambient intelligence and smart environments · 2011
An improved linear manifold learning method for object recognition is proposed. This method is called enhanced Locality Preserving Projections (LPP), and integrates two interesting properties: (i) being entirely parameter-free, and (ii) the mapped data are uncorrelated. The parameterless computation of the affinity matrix draws on the notion of meaningful and adaptive neighbors. Recognition tasks on five face data sets show a clear improvement over the results of a classical LPP. The proposed approach could also be applied to other categories of objects.