Online Graph-Based Change Point Detection in Multiband Image Sequences

TL;DR

This paper presents an online, graph-based method for detecting change points in multiband image sequences, leveraging graph signal processing and superpixel decomposition for efficient and localized anomaly detection.

Contribution

It introduces a novel online framework that uses graph modeling and superpixels to detect changes in multitemporal remote sensing images without prior change knowledge.

Findings

Effective detection and localization demonstrated in experiments

Scalable and computationally efficient approach

Utilizes graph signal processing and superpixel decomposition

Abstract

The automatic detection of changes or anomalies between multispectral and hyperspectral images collected at different time instants is an active and challenging research topic. To effectively perform change-point detection in multitemporal images, it is important to devise techniques that are computationally efficient for processing large datasets, and that do not require knowledge about the nature of the changes. In this paper, we introduce a novel online framework for detecting changes in multitemporal remote sensing images. Acting on neighboring spectra as adjacent vertices in a graph, this algorithm focuses on anomalies concurrently activating groups of vertices corresponding to compact, well-connected and spectrally homogeneous image regions. It fully benefits from recent advances in graph signal processing to exploit the characteristics of the data that lie on irregular supports. Moreover, the graph is estimated directly from the images using superpixel decomposition algorithms. The learning algorithm is scalable in the sense that it is efficient and spatially distributed. Experiments illustrate the detection and localization performance of the method.