Change Detection in Heterogenous Remote Sensing Images via Homogeneous Pixel Transformation

Abstract

The change detection in heterogeneous remote sensing images remains an important and open problem for damage assessment. We propose a new change detection method for heterogeneous images (i.e., SAR and optical images) based on homogeneous pixel transformation (HPT). HPT transfers one image from its original feature space (e.g., gray space) to another space (e.g., spectral space) in pixel-level to make the pre-event and post-event images represented in a common space for the convenience of change detection. HPT consists of two operations, i.e., the forward transformation and the backward transformation. In forward transformation, for each pixel of pre-event image in the first feature space, we will estimate its mapping pixel in the second space corresponding to post-event image based on the known unchanged pixels. A multi-value estimation method with noise tolerance is introduced to determine the mapping pixel using -nearest neighbors technique. Once the mapping pixels of pre-event image are available, the difference values between the mapping image and the post-event image can be directly calculated. After that, we will similarly do the backward transformation to associate the post-event image with the first space, and one more difference value for each pixel will be obtained. Then, the two difference values are combined to improve the robustness of detection with respect to the noise and heterogeneousness (modality difference) of images. Fuzzy-c means clustering algorithm is employed to divide the integrated difference values into two clusters: changed pixels and unchanged pixels. This detection results may contain some noisy regions (i.e., small error detections), and we develop a spatial-neighbor-based noise filter to further reduce the false alarms and missing detections using belief functions theory. The experiments for change detection with real images (e.g., SPOT, ERS, and NDVI) during a flood in U.K. are given to validate the effectiveness of the proposed method.

Unsupervised Image Regression for Heterogeneous Change Detection

“…The copula-based approach is an example of a method which maps data from one input image domain to the other. Another example is the more recent HPT method presented by Liu et al [9], which uses kernel regression on a sample of nearest neighbour pixels to set up mappings between the input domains. The regression function is learned with training data labelled as unchanged.…”

“…Furthermore, an error measure intrinsic to the RF process, the so-called out-of-bag error, can also be used to guide the tuning of the hyperparameters [19], [47]. In the case of HPT, guidelines on hyperparameter tuning are provided in [9] although automatic hyperparameter-optimization methods have not been developed so far. A general strategy may be based on cross-validation, although at the cost of increased computation time.…”

“…The homogeneous pixel transformation (HPT) method proposed by Liu et al [9] is a kernel regression algorithm [33] based on the K-nearest neighbours of each data point. This technique recalls the distance weighted averaging or locally weighted regression previously presented in [46], where many related aspects are also studied: possible kernels, distance measures, choices of the bandwidth, denoising techniques, and outlier detection.…”

“…The latter case invalidates the assumptions that the same physical quantities are measured, classes have always the same signatures, and data follow the same statistical behaviour [8]. Since traditional homogeneous CD techniques are unsuitable in this case, additional pre-or postprocessing steps are required [8], [9]. To meet this challenge, several kinds of methodologies have been proposed.…”

“…It is based on regression concepts and automatically generates the training dataset T for regression, which consists of M corresponding pixel pairs from areas in the two images pre-arXiv:1909.05948v1 [cs.CV] 7 Sep 2019 sumed not to be affected by changes. According to [2], [9], the provision of such training data by manual selection would not be a strong requirement, although it prompts user interaction. Nonetheless, besides being a costly and time-consuming operation, it also might lead to inaccurate information, especially when images are difficult to interpret [13].…”