Rock images classification using Principle Component Analysis and Spatial Frequency Measurement

摘要

Since the natural rocks have quite different textures even they are in the same class, it is very difficult and challenging task to classify each type of natural rocks. In this paper, we present a method to classify each type of rocks using the modified version of Spatial Frequency Measurement (SFM). In our approach, each type of color rock images are firstly transformed into two dimensional intensity features, obtained from the highest and lowest eigenvalues of the Principle Component Analysis (PCA). The highest and lowest eigenvalues are corresponded to the most and least significant feature components. Next, the textural contents of each component are measured using the modified version of SFM, which measures all overall activity level of each component in two directions including vertical, horizontal directions by shifting one by one pixel for two-neighborhood pixels in both direction. Before applying modified version of SFM, the edge detection operator, Sobel operator, is applied to the most significant component only. After applying the modified version of SFM to both components, two textural features are used to define each type of rock. In our experiments, we test our approach to classify on 14 different classes of rock textures, each class has 30 samples. From the results, we found that the scatter plots of each type of rock features are obviously grouped and stuck together in the same class while the different classes are clearly separated.