F-SIFT and FUZZY-RVM based efficient multi-temporal image segmentation approach for remote sensing applications

Abstract

Image segmentation plays a most important role in the remote sensing applications, for the efficient detection of the Earth surface. The main objective of the segmentation process is to modify and simplify the representation of an image into an easier form for efficient analysis. The performance of the image segmentation process reduces due to the occurrence of noise and disturbances in the image. Existing segmentation approaches suffer from the performance degradation in the segmentation accuracy owing to the quality of the acquired satellite image. To overcome these drawbacks, this paper proposes an efficient image segmentation process for the clear view of the multi-temporal satellite image. Gaussian Filter (GF) is used for filtering the image to remove the noises present in the image. PSO-Affine based image registration is applied for the extraction of the pixel points and registration of the multi-temporal image. Removal of cloud from the image is performed to get a clear view of the image. Feature extraction is performed by using the Fast-Scale Invariant Feature Transform (F-SIFT) approach. The feature points of the image are extracted to form the cluster including six different classes such as building area, road area, vegetation area, tree area, water area and land area. The classes of the cluster are recognized by using the Fuzzy-Relevance Vector Machine (F-RVM) algorithm. The proposed approach achieves better performance in the cloud removal and efficient image segmentation.