OBJECT RECOGNITION BY USING ARTIFICIAL NEURAL NETWORK: Turning point based shape recognition using NN

An artificial neural network (ANN) classifier for recognizing an object based on
their shapes is presented, regardless their position, orientation or size. To extract
features of an object, the significant point on the object known as comer or break
point is extracted and the object shape is approximated by connecting this extracted
break point with straight line. Shape features are associated with each segment
consisting of three successive break points, or any two lines in the approximated
shape. These features are the ratio between any two adjacent lines and the angle
between them. The extracted features are used as input the ANN. The neural network
configuration used in this project is multi-layer perceptron using back-propagation
learning algorithm. In this project two type of shape have been recognized by a MLP.
The network performance is evaluated by presenting several examples to the network
and determines the difference between the tested image and the original shape used in
the training, until the differences are minimized.