The purpose of this article is to shed light on the mechanism
and the procedures of a program that classifies an input face into
any of the six basic facial expressions, which are Anger, Disgust,
Fear, Happiness, Sadness and Surprise, in addition
to normal face.
This program works by apply PCA- principal component
analysis algorithm, which is applied of one side of the face, and
depends, on contrast to the traditional studies which rely on the
whole face, on three components: Eyebrows, Eyes and Mouth.
Those out-value are used to determine the facial feature array as
an input to the neural network, and the neural network is trained by
using the back-propagation algorithm. Note that the faces used in
this study belong to people from different ages and races.
Video image data can be analyzed and processed in many ways. This research explores the extent at which spiking neurons, which are designed along the Hodgkin-Huxley model, are suitable for this task. The simulations reported in this research
consid
er integrate-and-fire neurons constant and alternating input currents, as well as pixel-intensity driven inputs. Currently, the simulation software employs 64 independently operating spiking neurons that process image data taken every 25 ms. In order to define the response of these neurons, the experiments were done on 100 digital images which include different illuminations, contrast, and saturation situations. The results show that the integrate-and-fire-neuron is highly sensitive to the changes in the intensity of pixels if its parameters are properly set. So in many applications, such as "Saliency Maps", which highly depend on the intensity values of a set of pixels, a neural network made of this neuron will perfectly fit.
In this study we developed an adaptive model inspired by internal models in the cerebellum and this approach called Feedback Error Learning (FEL). FEL is the origin of Learning Feed-Forward Control (LFFC). It depends on Feedback Controller and Feed-F
orward Controller which is a Neural Network, and this Neural Network uses feedback controller output as training signal. We developed this approach to control a robot arm, and to balance inverted pendulum and to control bus suspension system. We developed this approach by adding a second Neural Network, and this new Neural Network uses FEL controller output as training signal. We simulate these systems by using Matlab and Simulink, and we find that this development improves control performance.
This paper introduces a system to recognize labels of time plans, where labels are
extracted from time plan. This labels are images, so spatial segmentation is used to extract
images of labels only. Size of images of labels are made same using medi
an's algorithm for
two purposes. The first one is to create database training for used neural networks. The
second is to recognizing's processing. Two methods of recognizing are dependent on using
neural networks technic: classification using perceptron network and recognizing using
back propagation network. Perceptron network is built to take image as input and to give
classification index as output for label. Then label is recognize dependent on stored table
of ASCII for label. Back propagation network is designed to recognize images for all
letters of English alphabet that are used in time plan. Results of research appear efficiency
of designed system to recognize labels of time plan from their images for both methods
after system had been applied on three time plans.
