|
Ultra fast cnn based Hardware Computing Platform Concepts for adas visual Sensors and Evolutionary Mobile RobotsBog'liq Alireza Fasih4.2
A survey of the related state-of- the art on image processing
based on ANN.
An artificial neural network is a mathematical model or computational model of biological
neural network. These are essentially a very simple mathematical model for mapping input
to output. The things that make ANN attractive are learning possibility. Means during the
learning phase, by giving a class of function system can find an optimum solution. We have
many different type of neural network such as Feedforward
Multi-layer Perceptron Neural
Network
(MLP),
Radial Basis function
(RBF), Kohonen
Self-Organizion Network
(SOM),
Hopefield Network,
Bi-directional Neural Network
(BRNN),
Associate Neural Network
(ASNN) and etc that are useful for different application. The most common type of artificial
neural network in the field of image processing is MLP. This model as it mentioned in
Figure 4-1, has many input, hidden layer and output. Each neuron in one layer is connected
the other neuron in another layer. Therefore we have fully connection between each layer
to another layer. Depending to the type of data and information we have to reshape them
for feeding them in our network. For example for loading an image into the network we
have to reshape the 2D picture in form of vector. Figure 4-2 has shown this concept. We
have to consider many issues before loading data to the network for training or in
operational mode. For image processing, in many cases we need to perform a color
conversion, normalization and quantization of values before feeding data into the network.
|
| |
