Rozpoznávání a třídění objektů podle tvaru

Object Sorting Based on Shape

článek ve sborníku ve WoS nebo Scopus

čeština

This paper presents modern science appears from the basis of Computer Vision and classifier of Synthetic Neural Network. A picture of watched object has to be taken in high quality with the best lighting and camera has to be situated vertically upon the object. Different camera positions does not assure exact results. Subsequently, image has to be transformed into binary image and purged from noise and other interference. Moments are used to describe separate objects in picture. Via the central moments and normed moments I count seven moments characteristic for each object to be identified. These moments are practically independent on rotation or changing scale of the object. They fluctuate only in a short spread. It is input to Neural Network, which is used as the classifier. The system of Back-propagation is used as the Neural Network with type of learning called learning with teacher. In my work, each letter of alphabet is used as the object to be identified. Further, I tried to identify object by using a tolerance. But as described before, the counting moments are not changeless. By the change of scale, moments fluctuate more than by rotation, thereafter if any value of identified object is out of the tolerance set, then we are not able to make the identification. Neural Network solve this problem. Enumeration of each moment and Neural Network is programmed in the Matlab 6.5 environment.

This paper presents modern science appears from the basis of Computer Vision and classifier of Synthetic Neural Network. A picture of watched object has to be taken in high quality with the best lighting and camera has to be situated vertically upon the object. Different camera positions does not assure exact results. Subsequently, image has to be transformed into binary image and purged from noise and other interference. Moments are used to describe separate objects in picture. Via the central moments and normed moments I count seven moments characteristic for each object to be identified. These moments are practically independent on rotation or changing scale of the object. They fluctuate only in a short spread. It is input to Neural Network, which is used as the classifier. The system of Back-propagation is used as the Neural Network with type of learning called learning with teacher. In my work, each letter of alphabet is used as the object to be identified. Further, I tried to identify object by using a tolerance. But as described before, the counting moments are not changeless. By the change of scale, moments fluctuate more than by rotation, thereafter if any value of identified object is out of the tolerance set, then we are not able to make the identification. Neural Network solve this problem. Enumeration of each moment and Neural Network is programmed in the Matlab 6.5 environment.

Neuronove sitě; rozpoznávání obrazu

Neural Network; Back-propagation; Computer Vision

TOFEL, P.

2007

19. 1. 2007

VUT Brno UFYZ

Brno

978-80-7355-075-2

NOVÉ TRENDY V MIKROELEKTRONICKÝCH SYSTÉMECH A NANOTECHNOLOGIÍCH

První

121

125

5

Knihovna UFYZ