This letter proposed an advanced superpixel-based Markov random field (MRF) model for unsupervised change detection. First, simple linear iterative clustering was implemented to the difference image to get a superpixel map. The initial change map is then obtained using fuzzy c-means by clustering the superpixel map. Finally, the initial change map of superpixel is iteratively refined by improved superpixel-based MRF that incorporated the spatial attraction model. It considers the spatial information of adjacent superpixels to complete the miss-detection superpixels and eliminate false-alarm superpixels. The experimental results over two data sets demonstrate the effectiveness of the proposed method.

An improved graph-cut-based change detection method is proposed in this paper to make full use of the spectral and spatial information from multispectral remote-sensing images. The proposed method detects changes by minimizing the graph-cut energy function. The energy function consists of change feature energy and image feature energy. The two features are generated based on spectral information and spatial-context information, respectively. Change feature energy item is calculated from the change vector, which uses the spectral information to detect changes. Image feature energy item is obtained by calculating the similarity of the texture features between neighbouring pixels. The image feature energy item uses spatial information to refine the contours of change detection results and remove false alarms (FA). A novel energy function is proposed to quantify the spatial-context information and measure the difference information between multispectral images. Finally, the max-flow/min-cut method is employed to produce the final change map by minimizing the energy function. The experiments carried out on medium- and high-resolution images demonstrate the robustness and effectiveness of the proposed method. This study provides a new perspective for incorporating spectral and spatial information in change detection.