期刊名称：ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences
印刷版ISSN：2194-9042
电子版ISSN：2194-9050
出版年度：2000
卷号：XXXIII Part B5/1
页码：720-726
出版社：Copernicus Publications
摘要：In order to describe a temporal image sequence, global information about each separate image and their interrelations is required. Traditional moments, as applied to single images, are designed to capture global information about an object, but are unable to describe temporal changes. In this paper a new global statistical description is proposed - velocity moments, to compress and characterise information from temporal (moving) features. These new velocity moments are based on traditional moment theory. The sequence of images to be characterised is treated as a single entity and the moments aim to describe how different images are interrelated. The new method shows promising recognition properties when used with simple synthetic sequences of moving shapes. Due to their structure they exhibit similar properties to traditional moments when reconstruction is applied using a technique called moment matching. The noise performance of the velocity moments and a set of traditional Hu invariant moments are evaluated. Experimental results show that the velocity moments exhibit a higher resilience to perimeter noise, which suggests that they are better suited to describing real-world, extracted temporal features. As such, a new statistical method has been proposed to statistically describe moving shapes, which combines information about the shape's structure with information about its movement
关键词：Moments; Moment Reconstruction; Temporal Desciption;Statistical Description; Image processing
