SUMMARY
The success of the tobacco leaf classification process is very dependent on the extraction of tobacco leaf features. Several stages of digital image processing can improve the ability to identify the best quality tobacco automatically through extracting leaf texture features. This study aims to apply the leaf texture feature extraction system using the Discrete Cosine Transform method. Classification results measure the accuracy of the success of the system in extracting the best texture features. The classification of tobacco leaves requires extensive knowledge and complex terminology, even professional graders require significant time in this field for mastery of the subject. This is because tobacco leaves are usually considered to have characteristics that are useful for identification of tobacco quality where the extraction of appropriate features through leaf images can be considered a research problem that plays an important role for classification. The proposed research aims to find a suitable extraction model for obtaining color features through YCbCr color space conversion and tobacco leaf texture obtained from the transformation of the Discrete Cosine Transform frequency space. On classification stage in this research uses the maximum likelihood method. The trial results show an accuracy of success in the classification of tobacco leaves by 90% through the extraction of 12 features.