13-06-2012, 04:44 PM
3D FACE RECOGNITION USING 3D RPROP ALGORITHM
3D FACE RECOGNITION.ppt (Size: 962.5 KB / Downloads: 58)
OBJECTIVE
Development of 3D RPROP algorithm as a faster training method for 3D Artificial Neural Networks and using that for 3D face recognition, by using a 3D co-ordinate pattern obtained for each face from 3D VRML Database.
DATABASE USED…
We have used GavabDB[1],
It contains 427 three-dimensional facial surface images corresponding to 61 individuals (45 male and 16 female), and there are 7 different images per each person.
Each image consists in a three-dimensional mesh representing a face surface.
Every mesh (captured 3D image) is composed of points of the facial surface and their connections, forming cells of four non-coplanar nodes each one.
The coordinates (x, y, z) of their 3D points are referred to a coordinate origin placed in the scanner during the capture time. Each mesh has been exported to a VRML format file.
VRML..
The codes for 3D worlds are written in VRML 1.0 or VRML97 format.
Our Database uses VRML 1.0 for describing 3D information face.
Virtual Reality Modeling Language is the language used to display three-dimensional objects with a VRML viewer.
PRELIMINARIES
ARTIFICIAL NEURAL NETWORKS
Computational analogue of human neurons.
Particular inputs are mapped to particular outputs.
Squashing Activation functions are used e.g. log-sigmoid, tan-sigmoid etc.
Complex Relationships among Input-Output pairs are established through weight changes.
RESILIENT BACKPROPOGATION
Sigmoid functions are characterized by the fact that their slopes must approach zero as the input gets large.
The gradient can have a very small magnitude and, therefore, cause only small changes in the weights and biases, even though the weights and biases are far from their optimal values.
Purpose resilient backpropagation (Rprop) algorithm is to eliminate these harmful effects of the magnitudes of the partial derivatives.
Only the sign of the derivative can determine the direction of the weight update; the magnitude of the derivative has no effect on the weight update.
The size of the weight change is determined by a separate update value. The update value for each weight and bias is increased by a factor delt_inc whenever the derivative of the performance function with respect to that weight has the same sign for two successive iterations.
Update value is decreased by delt_dec the sign of derivative changes for two successive iterations.
WORKING ALGORITHM
3D VECTOR BACKPROPOGATION ALGORITHM
It uses a novel neuron model whose input and output signals, and threshold Values are all 3D real valued vectors, and whose weights are all 3D orthogonal matrices.
3DV-BP network has the ability to learn 3D affine transformations.