Abstract:
This paper presents a new object verification framework to choose the correct building hypothesis in natural images. Compared to conventional approaches that extract the single feature, and assume little knowledge, the proposed approach extracts edge features and line-segment features, and turned object verification into a conditional probability when conditioned upon an object hypothesis. Based on the Bayesian theory, the prior knowledge can be converted into a series of prior probabilities to compute the maximum a posteriori estimate, so a new approach to object verification is presented. Experiments on the natural image sets demonstrate that the proposed approach can yield substantial improvements over the traditional approach on the performance of recognition.
Key words:
natural image,
observation,
object verification,
Bayesian
摘要: 提出了一种对自然图像中候选的建筑物目标进行验证的方法。与传统的提取单一图像特征,利用少量先验知识进行验证的方法相比,该方法提取图像的边缘特征和短线段特征,通过建筑物图像中特征和特征分组的观察,将目标验证转化为给定候选目标的条件概率问题。利用贝叶斯理论,将建筑物目标的先验知识表现为一系列先验概率并计算后验概率的值,从而给出了一种新的目标验证方法。利用拍摄的自然图片进行实验表明:与传统的方法相比,该方法的识别性能有了一定程度的提高。
关键词:
自然图像,
观察,
目标验证,
贝叶斯
CLC Number:
JIN Tai-song; LI Cui-hua; LIU Ming-ye. Approach to Building Object Verification in Natural Images[J]. Computer Engineering, 2007, 33(16): 4-6.
金泰松;李翠华;刘明业. 一种自然图像中的建筑物目标验证方法[J]. 计算机工程, 2007, 33(16): 4-6.