遥感卫星技术的发展使得高空间分辨率的遥感图像得到了更多的应用。但是，高空间分辨率遥感图像常常具有高的类内方差，由此限制了许多遥感信息提取方法的性能。为解决此类问题，图像中像素间的空间约束成为研究热点，且陆续产生了一些研究成果。但整体上，这些成果缺少联系和系统性。鉴于此，本文基于近二十年来一百余篇的相关文献，对现有空间约束流程、应用场景和方法进行了归纳与总结，对各类方法进行了原理解释及优缺点的比较。最后，分析了空间约束方法的发展趋势，列举了空间约束研究存在的可能的不足，提出了建议参考。

The problem of a high intra-class variance arose in the very high-spatial-resolution remote sensing images (VHR) and limited the performance of many remote sensing information extraction methods. To solve such problems, the spatial constraint (SC) among pixels of images has become a hot topic, and produced many research results, but lack of association and systematicness as a whole. This paper reviewed and summarized more than one hundred related literatures published in the past two decades, so as to provide references for further research on information extraction in VHR. There are four sections in this paper. First, SC process is divided into three stages: mining and expression of spatial information, and construction of SC, was introduced in detail. In generally, the main sources of spatial information are the neighborhood of pixels, imaging relation, prior knowledge. The expression of spatial information included mean, median, extreme, azimuth order, etc. The construction methods of SC included objective function, energy function, discriminant function, etc. Second, SC applications are divided into six scenarios: image matching, image segmentation, target detection, image classification, change detection, and others, the implementation methods and characteristics of main application scenarios are summarized. The way of SC is closely related to specific application. For example, SC is mainly used to build the descriptor and transformation in image matching, implemented by model constraint, graph construction in space, and objective function in image segmentation, target detection and image classification, and emphasized on the neighborhood between pixels and prior knowledge in change detection. The common features in these scenarios are to develop a robust, unique and representative descriptor via the geometric space information to solve the specific problems in the images. Third, SC methods are divided into six types: local templates, auxiliary reference, graph construction in space, model constraints, rule constraints, and others, and are summarized in the paper according to the implementation and principles. The advantages and disadvantages of the first five methods are compared. The results showed that the different SC methods presented the different usability in different application scenarios. (1) A local template uses the spatial information of the neighborhood and can obtain more stable information expression, so is suitable for many application scenarios, especially image classification. (2) The point constraint in auxiliary reference uses the spatial relations between feature points and often appears in image matching, line constraint focuses on the connection between the target and the linear object and is suitable for extraction of man-made objects, and surface constraint is of spatial extensibility and suitable for target detection. (3) Graph construction in space uses the multidimensional spatial information intuitively and effectively and suitable for classification in hyperspectral images. (4) Model constraints are of generalization in applications, but dependent on the specific mathematical expression. (5) Rule constraint specifies the professional applications and often are used in image classification and change detection. Fully analysis and consideration of application scenarios and specific problems still are the preconditions of SC effectively. In the last, this paper presents the development trend and the possible shortcomings in the spatial constraint research, and specific suggestions for works in the future.