基于模型约束的不完整图象的加标方法

An Approach for Labeling Imperfect Image Based on Model Constraints

加标是计算机视觉研究的内容之一。用传统的加标方法处理不完整图象遇到许多问题。本文提出的方法利用了模型知识在不同处理阶段所产生的不同约束作用对不完整图象中的不精确特征加以限制和排除，逐渐完善图象的描述，以最终达到与模型的一致性。高阶约束的引入和最大约束特征的利用是该方法所采取的两项措施，用来产生附加约束，减少对图象特征的歧义性解释。

Labeling is one of research area of computer vision. The problems occurwhen traditional algorithms are used for labeling imperfect image. In this paper a model constraints-based approach for labeling imperfact image is proposed. In order to limit and discard the uncertain feature, refine the image, and construct a consistent description with model, in proposed method, the model knowledge are divided into different levels and used in different phases during label processing. The effection of high order constraints and most-constrainted-feature are also disccused which can provide the additional constraints for labeling to reduce the ambiguity.