Object Predetection Based on Kernel Parametric Distribution Fitting

TL;DR: This work proposes efficient and accurate variants of multimodal distribution fitting techniques convenient for object predetection in images based on parametric distribution fitting in the RKHS space induced by a positive definite kernel.

Abstract: Multimodal distribution fitting is an important task in pattern recognition. For instance, the predetection which is the preliminary stage that limits image areas to be processed in the detection stage amounts to the modeling of a multimodal distribution. Different techniques are available for such modeling. We propose a pros and cons analysis of multimodal distribution fitting techniques convenient for object predetection in images. This analysis leads us to propose efficient and accurate variants over the previously proposed techniques as shown by our experiments. These variants are based on parametric distribution fitting in the RKHS space induced by a positive definite kernel.