ABSTRACT
Change detection is a very ancient topic for study. In decades of research, two main frames make contributions to its development: post-classification and direct comparison. In the late 1990s, post-classification was the main research subject. Later, direct comparison began to attract much more attention and became the hot research direction for its simplicity, comprehensible, and high efficiency. Direct comparison consists of two main steps: difference image building, difference image analysis, and post-processing. The method of building difference image includes gray difference method, gray ratio method, and texture method and so on. For the subsequent analysis techniques, thresholding method is employed as the principal stage or an inserted step in the processing of parameter optimization. Most of the threshold methods are based on the minimization or maximization of one criterion function, such as minimum error thresholding method with Gaussian model [1, 5] or generalized Gaussian model [3, 4].These methods find the so-called optimal threshold through the minimization or maximization of the entropy of one probabilistic model. The times of iterative optimization is equal to the max gray level for the criterion function of all the gray level in the image need to be computed.To decrease the operational times of threshold optimization, one iterative approximation method of minimum error thresholding algorithm is proposed to analysis the difference image. In addition, different kinds of difference images show up change information in different styles. One dynamic fusion strategy is employed to construct difference
image, which utilize different styles of change information.
