An image selection method for tobacco leave grading based on image information

Abstract

In order to select grading images of tobacco leaves, the reflecting light images of all tobacco leaf samples are gathered under the condition of seven color light sources. Then for the each color light, the self-information (SI) of each sample image is calculated, and the mean SI of all images corresponding to one grade of tobacco leaf samples (MSI) is also calculated. Furthermore, the mutual information (MI) between two sample images corresponding respectively two grades of tobacco leaf samples is calculated, and the mean MI between arbitrary two grades of tobacco leaves images (MMI) is also calculated. It is to be found that the ratio of the MMI to sum of the two MSI which correspond to two grades of samples can be employed to select the grading images, and the sample images obtained under the condition of the cyan light are selected for tobacco leaves grading. At the same time, 18 kinds of color features and 13 kinds of texture features are extracted from all sample images, and Fisher discriminant analysis (FDA) is used to distinguish these samples. The FDA results indicate that the grading accuracy based on cyan image is the highest. This is in accord with the above selection result. Therefore, the proposed selection method is effective and independent of features extracted from sample images.