Multifeature based automatic annotation of satellite images

Abstract

The remote sensing images have been widely used in hazard assessment, oceanic monitoring, natural resources management, etc. in the earth monitoring technologies. An image annotation is regarded as insertion of a keyword or tag, formulate color overlay on an image. Researchers developed annotation techniques for effective retrieval of information from the massive image dataset. In earlier days, manual annotation is much preferred for information retrieval process. But, it requires more time, a large quantity of labor and different idea generations etc. Hence, it is not preferred for satellite image annotation. In view of this, most of the researchers prefer an automatic annotation of images. To develop a multi feature based automatic annotation of satellite images with high accuracy and less computational cost, this research contributes three different automatic annotation methodologies. In the first Method, Discrete Wavelet Transform (DWT) based Linear Binary Pattern (LBP) features are defined for better discrimination of the satellite images regions. newlineAnd sum up the different kernel values of Multiclass Support Vector Machine (M-SVM) to improve the performance of annotation. And in the second Method, to improve the accuracy of the annotation process by hybrid classifier is used. For that purpose, Random Forest based Probabilistic Neural Network (rf-PNN) is implemented along with extra textural and color descriptors. And in the last Method, Adaptive Neuro-Fuzzy Inference System (ANFIS) is implemented with fusion of LBP, Texture and color descriptors. These methodologies have been experimented with some of benchmark datasets namely AID, UC-Merced and WHU-RS19. newline