Development of new feature extraction method for machine vision based measurement and quality inspection in conveyed objects

In this thesis, proposed a new feature extraction method for machine vision based measurement and quality inspection in conveyed object. In the scope of the study, the most commonly used methods in the literature were investigated and find out texture analysis methods have been used frequently. So the study is based on texture analysis. Two of the most common methods used in quality control applications are local binary pattern and gray level co-occurrence matrices, and Gabor Wavelet Filter is often used in recent studies. Within the scope of the study, application of these three methods were compared using two different databases. One of these databases consists of surface defects of rolled steel strips and the other database consists of flawed and perfect fabric images. Then a study was conducted to develop the local binary pattern method. In the next section, a new feature extraction method is proposed for the GLCM method. In the study, both the results obtained when using the proposed method alone and the results obtained with the features used in Chapter 3 are given. In the last section, accuracy rates were increased by using property selection method. It was seen that the new method proposed for GLCM in feature selection process increased the accuracy rates. After the feature selection process, the accuracy rates were obtained higher than the 93% in the NEU database and 90% in the fabric database for all directions and two separate distances. The studies developed within the scope of this master thesis were supported by the SAN-TEZ project number 0743.STZ.2014 (112D021) by TUBITAK and the Ministry of Science, Industry and Technology.