Abstract:
In this paper, a hybrid model is introduced which combines
a linear regression model in parallel with a nonlinear regression
model such as the Modified Probabilistic Neural Network
(MPNN). This model provides a first order approximation
of the underlying mechanism using linear regression,
and then use the MPNN to capture the local details of interest.
This model allows the selected data regions of interest
be modeled more accurately by a nonlinear compensator
while the rest of the data regions are approximated by a linear
regression model. The experiment on surveillance image
modelling shows that the proposed model achieves improved
performance over conventional methods such as MultiLayer
Perceptron (MLP) or Volterra Filter based modelling.
