Segmentation of Remote Sensing Images Using Similarity-Measure-Based Fusion-MRF Model

Abstract

Abstract-Classifying segments and detecting changes in terrestrial areas are important and time-consuming efforts for remote-sensing image analysis tasks, including comparison and retrieval in repositories containing multi-temporal remote image samples for the same area in very different quality and details. We propose a multi-layer fusion model for adaptive segmentation and change detection of optical remote sensing image series, where trajectory analysis or direct comparison is not applicable. Our method applies unsupervised or partly supervised clustering on a fused image series by using cross-layer similarity measure, followed by a multi-layer MRF segmentation. The resulted labelmap is applied for the automatic training of the single layers. After the segmentation of each single layer separately, changes are detected between the single label-maps. The significant benefit of the proposed method has been numerically validated on remotely sensed image series with ground-truth data.

Extreme anomaly event detection in biosphere using linear regression and a spatiotemporal MRF model

“…The dependency between adjacent pixels can be modeled by conditional probabilities within a neighborhood system. Sziranyi and Shadaydeh (2014) proposed a multi-layer fusion MRF model for change detection in multi-temporal optical images. This method has been further improved in Shadaydeh et al (2017) to deal recursively with time series of images.…”

“…To this end, the obtained Mahalanobis distance over the entire study area is treated as a time series of images. We use an adaptation of the multi-layer fusion MRF classification model presented in Sziranyi and Shadaydeh (2014) and Shadaydeh et al (2017) for the classification of this Mahalanobis distance images into three classes, intense anomaly, possible anomaly and normal.…”

Multilayer Markov Random Field models for change detection in optical remote sensing images

“…PCC methods (Liu and Prinet, 2006;Castellana et al, 2007;Zhong and Wang, 2007;Szirányi and Shadaydeh, 2014) segment first the input images into various land-cover classes, like urban areas, forests, plough lands etc. In this case, changes are obtained indirectly as regions with different class labels in the different time layers.…”

“…MRFs are able to simultaneously embed a data model, reflecting the knowledge on the measurements; and prior constraints, such as spatial smoothness of the solution through a graph based image representation, where nodes belong to different pixels and edges express direct interactions between the nodes. Although a number of the corresponding MRF based state-of-the art models deal with multispectral (Bruzzone and Fernandez-Prieto, 2002;Ghosh et al, 2007;Xu et al, 2012;Chen and Cao, 2013;Ghosh et al, 2013;Subudhi et al, 2014) or SAR (Melgani and Serpico, 2003;Bazi et al, 2005b;Carincotte et al, 2006;Gamba et al, 2006;Martinis et al, 2011;Wang et al, 2013;Baselice et al, 2014) imagery, the significance of handling optical images is also increasing (Zhong and Wang, 2007;Moser et al, 2011;Szirányi and Shadaydeh, 2014;Hoberg et al, 2015).…”

Segmentation of high resolution remote sensing images by combining hidden Markov random field model and fuzzy c-means at the region level

“…It aims at partitioning an image into disjoint meaningfully homogeneous regions. In the past decades, many different image segmentation techniques have been proposed [1,2,3,4]. Among them, those based on fuzzy c-means (FCM) [5] and Markov random field (MRF) models [6] are the most popular ones.…”