基于目标形状的图像检索

Image retrieval based on object shape

针对已有的基于形状的图像检索中目标形状描述方法的不足对其进行改进。首先对目标图像进行一系列预处理,得到图像的外部轮廓,利用改进的霍夫变换提取目标轮廓的线性特征;然后引入成对几何特征即有向相对角和有向相对位置来描述图像的形状;最后利用直方图相交算法衡量图像特征间的相似度。实验证明,利用本文改进的方法所描述的形状属性来检索数据库中的图像具有较高的效率。

This paper focus on improving the description methods of object contour in contour-based image retrieval. In order to get the external contour of image, a series of image pretreatment are carried out. Then, the modified Hough transform is used to extract linear features of object contour. It also uses the pairwise geometric attributes which are the directed segment relative angle and directed relative position to describe image shape. Finally, histogram intersection algorithm is used to measure the similarity between the image features. Expements conclude that a relation histogram of pairwise segment attributes presents a very efficient way of indexing into large databases.