Noisy satellite image segmentation using statistical features

Abstract

Satellite image segmentation is a principal task in many applications of remote sensing such as natural disaster monitoring and residential area detection and especially for Smart cities, which make demands on Satellite image analysis systems. This type of image (satellite image) is rich and various in content however it suffers from noise that affects the image in the acquisition. The most of methods retrieve the textural features from various methods but they do not produce an exact descriptor features from the image and they do not consider the effect of noise. Therefore, there is a requirement of an effective and efficient method for features extraction from the noisy image. This paper presents an approach for satellite image segmentation that automatically segments image using a supervised learning algorithm into urban and non-urban area. The entire image is divided into blocks where fixed size sub-image blocks are adopted as sub-units. We have proposed a statistical feature including local feature computed by using the probability distribution of the phase congruency computed on each block. The results are provided and demonstrate the good detection of urban area with high accuracy in absence of noise but a low accuracy when noise is added which yields as to present a novel features based on higher order spectra known by their robustness against noise.