基于角点特征的图像检索新方法

A NOVEL METHOD FOR IMAGE RETRIEVAL BASED ON CORNER FEATURES

多数基于角点的图像检索方法都是基于角点颜色特征提出的,并没有考虑特征显著性。针对此,对角点的多种特征进行实验分析,利用其中较为显著的两种特征,提出一种基于角点的特征进行图像检索的新方法。算法构建角点的广义灰度共生矩阵作为局部纹理特征;同时提取角点周围区域的Hu不变矩作为局部形状特征。最后对纹理特征和形状特征进行加权运算,将其加权结果作为检索特征实现检索。实验结果表明,选择合适的权值,结合角点的两种特征能实现更好的检索效果。

Most of corner-based image retrieval methods are brought forth based on corner colour features,but do not consider the prominence of the feature. In view of this,we make experimental analyses on a variety of features of the corner points,and propose a novel algorithm by using two more prominent features among them,which retrieves images based on corner features. The algorithm constructs the generalised gray level co-occurrence matrix of corner points as the local texture feature; at the same time it extracts Hu invariant moments of the area around the corner point as the local shape feature. Finally,it makes weighted operation on texture feature and shape feature,and uses the weighted result as the retrieval feature to implement retrieval. Experimental result shows that while selecting the appropriate weight,the retrieval can achieve a better result by combining the two features of corner points.