A new regional shape index for classification of high resolution remote sensing images

Based on the object-oriented method, this paper presents a new regional shape index (RSI). RSI is a regional feature which measures the gray similarity distance within region in every direction. Firstly, the original image is segmented to obtain small regions. Then the center in each region is calculated, and the distance is calculated from the center of each region within image to boundary of each region in every direction. Finally, the results by RSI are compared with some textural features extracted using Gray Level Co-occurrence Matrix (GLCM), Local Binary Patterns (LBP), Regional Gray Level Co-occurrence Matrix (R-GLCM), Regional Local Binary Patterns (R-LBP). Experiments are conducted on high spatial resolution remote sensing image of Washington DC obtained by HYDICE and texture synthesis image confirm that the proposed method is feasible and effective. These experiments demonstrate the classification approach based on RSI feature results in higher classification accuracy than other methods. In a word, classification approaches based the regional level feature results, such as RSI, R-GLCM and R-LBP in higher classification accuracy than those approaches that consider pixel-wise feature, such as GLCM and LBP.