[Paper Review] Detecting Changes Between Optical Images of Different Spatial and Spectral Resolutions: a Fusion-Based Approach

Vinicius Ferraris, Nicolas Dobigeon+2 more|Open Archive Toulouse Archive Ouverte (University of Toulouse)|Sep 20, 2016

Remote-Sensing Image Classification26 references4 citations

TL;DR

This paper proposes a novel fusion-based framework for unsupervised change detection between optical images with mismatched spatial and spectral resolutions, such as panchromatic/multispectral and hyperspectral images. By fusing low-resolution spatial and low-resolution spectral images into a high-resolution pseudo-image, predicting images matching each input's resolution, and applying change detection to matched pairs, the method achieves superior detection accuracy compared to conventional approaches, with AUC up to 0.991 in real-data experiments.

ABSTRACT

Change detection is one of the most challenging issues when analyzing remotely sensed images. Comparing several multi-date images acquired through the same kind of sensor is the most common scenario. Conversely, designing robust, flexible and scalable algorithms for change detection becomes even more challenging when the images have been acquired by two different kinds of sensors. This situation arises in case of emergency under critical constraints. This paper presents, to the best of authors' knowledge, the first strategy to deal with optical images characterized by dissimilar spatial and spectral resolutions. Typical considered scenarios include change detection between panchromatic or multispectral and hyperspectral images. The proposed strategy consists of a 3-step procedure: i) inferring a high spatial and spectral resolution image by fusion of the two observed images characterized one by a low spatial resolution and the other by a low spectral resolution, ii) predicting two images with respectively the same spatial and spectral resolutions as the observed images by degradation of the fused one and iii) implementing a decision rule to each pair of observed and predicted images characterized by the same spatial and spectral resolutions to identify changes. The performance of the proposed framework is evaluated on real images with simulated realistic changes.

Motivation & Objective

• Address the lack of robust, flexible, and scalable change detection methods for optical images acquired by different sensor types with mismatched spatial and spectral resolutions.
• Overcome limitations of conventional methods that assume identical sensor characteristics or rely on suboptimal resampling and interpolation.
• Develop a unified framework applicable to emergency scenarios where only heterogeneous image pairs are available.
• Enable accurate change detection by exploiting the complementary information in low-spatial-resolution and low-spectral-resolution images.
• Provide a physically grounded, unsupervised approach that does not require labeled change data or prior knowledge of change locations.

Proposed method

• Fuse two input images—one with low spatial resolution but high spectral resolution (e.g., hyperspectral), the other with low spectral resolution but high spatial resolution (e.g., panchromatic)—into a single high-resolution, high-spectral pseudo-image using a physics-based image fusion model.
• Predict two images from the fused pseudo-image: one matching the spatial resolution of the high-spatial-resolution input and another matching the spectral resolution of the high-spectral-resolution input, using degradation models.
• Apply standard change detection techniques (e.g., CVA, spatially regularized CVA) to each pair of observed and predicted images with identical spatial and spectral resolutions.
• Use a spatially regularized change vector analysis (sCVA) to improve detection accuracy by incorporating local spatial context.
• Formulate the fusion step as a constrained optimization problem that preserves spectral and spatial consistency.
• Validate the framework through real data experiments with simulated changes, using AUC and normalized distance as performance metrics.

Experimental results

Research questions

• RQ1 Can a unified framework be developed to perform change detection between optical images with different spatial and spectral resolutions, such as hyperspectral and panchromatic images?
• RQ2 How can the complementary information in low-spatial-resolution and low-spectral-resolution images be effectively combined to improve change detection accuracy?
• RQ3 To what extent does the proposed fusion-based prediction strategy outperform conventional resampling and interpolation techniques in change detection?
• RQ4 Does incorporating spatial regularization in the change detection step significantly improve detection precision and reduce false positives in heterogeneous image pairs?
• RQ5 Can the proposed 3-step procedure (fusion, prediction, detection) be generalized to other image modalities beyond optical remote sensing?

Key findings

• The proposed method achieved an AUC of 0.991192 on the HR-MS scenario and 0.991545 on the HR-PAN scenario, significantly outperforming the worst-case approach (AUC 0.937878 and 0.971988, respectively).
• The high-resolution change detection map ($\hat{\mathbf{D}}_{\mathrm{HR}}$) produced by the framework demonstrated superior detection rate and precision, with a normalized distance of 0.959296, indicating high accuracy.
• Spatial regularization in the CVA method (sCVA) further improved performance, with AUC reaching 0.989493 and normalized distance 0.958996 in the HR-PAN scenario.
• The framework outperformed the standard approach of resampling both images to a common resolution, which led to a significant drop in performance, especially in spectral resolution degradation.
• Visual results in Fig. 6 and Fig. 7 confirmed that the estimated HR CD map was sharper and more accurate than the LR maps, with fewer false positives and better localization of small changes.
• The method demonstrated robustness across different change sizes, from 1×1 to 61×61 pixels, maintaining high detection fidelity.