الفهرس 
IntroductionOnly 14 pages are availabe for public view
 |  | from | 214 |  |  |
| | | from | 214 | | |
Abstract
ABSTRACT The Most important tool in quality control is the testing methods and instruments. The recent developments in the field of quality control lies in the introduction of new techniques for production monitoring known as “on- line quality control”. The most promising technique in this manner is the computer vision. The computer vision is more objective methodology that can be an excellent alternative to replace the visual inspection of human workers. Computer vision is a relatively new technology that combines computers and video cameras to acquire, analyze, and interpret or classify images in a way that parallels human vision. For this purpose, neural perceptron is a very efficient tool. This work presents an approach to detect the defects of the products. It aims to build a vision system that is able to detect and classify some of the famous carpet defects. The test materials consist of four different tufted carpet products. They are referred to as series A, B, C, D. Each class contains a normal (non – defective), and defected samples The defects are studied by statistical analysis of image matrices for tonal features (Mean, Variance, Skewness, Kurtosis, Entropy),and texture features( A.S.M., Contrast, Correlation, Entropy). The considered defects are: missing pile, higher pile, lower pile, slobs and knots; Image files are manipulated in three different formats: true color, 256 color and 256 gray levels. Textural features were evaluated by computing spatial gray level dependence (SGLDM) and gray level difference (GLDM) statistics. An artificial-neural–network was used for detecting the carpet defects. A single hidden- layer neural network trained by using the perceptron algorithm, mainly depends on a software program. To provide the data required to train the networks a visual basic program was designed. ANOVA and LSD were made to choose the significant features from all produced features. Several neural networks were built and trained with changing both of the network structure and the number of training samples. The best of these networks were selected to be used for the detection of defects.