基于内容图像检索中针对显著方向的有向滤波器

目前有向滤波器主要适用于广义图像域,而不适用于狭义图像域.针对该问题,引入统计方法和模糊理论,在每个方向对狭义图像域子集进行聚簇分析,根据分析结果确定隶属度,把隶属度作为该方向纹理特征向量的权重,并且根据隶属度确定k显著方向.从k显著方向获取纹理特征组成纹理特征向量,利用欧几里德距离确定纹理特征向量间的相似性.实验表明,由π,π2,0,32π方向计算的纹理特征向量在匹配精度上优于从0,π4,π2以及34π计算的,而且该有向滤波器适用于狭义图像域·

Orientational Filter Based on Dominant Directions in Content-Based Image Retrieval

The orientational filter is an integral part of content-based image retrieval, which is now mainly applicable to general image field instead of particular image field. Statistical method and fuzzy set theory are introduced in the clustering analysis which is made for the subsets of the particular image field in each and all directions. The degree of membership is computed according to the results of clustering analysis and used as the weight of the texture characteristic vector in a certain direction, and k dominant directions are determined correspondingly according to the degree of membership. Texture characteristics are thus computed from the k dominant directions and used to form texture characteristic vector. Similarity measure between texture characteristic vectors is computed by means of Euclidean distance. Experiments showed that the texture characteristic vector obtained from π, π/2, 0, 3/2π directions is better than that obtained from the computation of 0, π/4, π/2 and 3/4π in matching precision. Moreover, such an orientational filter is applicable to particular image field.