"Scene Exploration Using Bayesian and Neural Networks",
is a project carried out as a part of "3D Image Analysis and Synthesis"
at the Graduate Research Center of Chair of Pattern Recognition and is
supported by German Research Foundation. The present project may be
classified into one of the advanced fields of image processing
and finds its application where challenge of the machine
perception in complex scenes and
work environments is sought.Exploring scenes using Bayesian nets (BNs) is based on the idea of
performing an active knowledge based search on images, unlike conventional
visual recognition algorithms. During the indirect search of images,
a sample set of training images from different classes is available
right at the onset of an experiment and the nature of the class
to be searched is unknown. Usually a recursive search
for objects in an image, belonging to all classes is performed using a
conventional object recognition system and the Bayesian approach, the
goal of the present research work, can obviate this.
The search of objects in an image by BNs can be confined
only to a specific class or a set of classes. Our initial results have
proved that if structural relationships are rightly established
between the constituent objects of an image,
searching scenes using BNs is quite effective. However the BN structure
and the parameters are manually specified in our initial experiments.
Encouraged by the initial results, obtained by manual specification of
structure and parameters to the BNs, presently we are applying
Gaussian Mixture and QMR Models for object recognition.
By applying these models for object recognition obviates the manual
specification of parameters and BN Structure. EM Algorithm is employed
to compute the A Posteriories for generic object recognition.