基于内容的卫星云图检索方法

为了有效地管理和利用大规模卫星云图数据库,需要为其增加检索功能,尤其是基于内容的检索.探讨了将基于内容的图像检索(CBIR)技术应用于卫星云图检索,选择使用了颜色、纹理和形状特征方法.颜色特征包括:灰度直方图和灰度聚合向量;纹理特征从小波子图提取,采用直方图拟合法和分块直方图拟合法;形状特征提取采用了新提出的一种专门针对云目标的"可变形模型"方法——圆形基元法.对每种特征定义了图像距离或者相似度,进行了检索比较实验.结果表明,CBIR用于卫星云图检索是可行的,形状特征方法因融入了领域知识而具有较突出的检索效果,但效率仍需提高.

CBIR methods in satellite cloud image retrieval

In order to ef ficiently manag e and make use o f large-scale satellite cloud image col lect io ns, it is useful and essent ial to build a database and to equip it w ith ret riev al funct io ns, especially content-based retr ieval ones. Content-Based Image Ret rieval ( CBIR) techniques were applied to satellite clo ud imag es. A series of metho ds were used to ex tract colo r, texture and shape featur es. Color features used w ere g ray level histo gram and g ray level coherence v ector. T ex tur e features w ere ex tr acted fr om w avelet subbands using tw o methods: histogram fit t ing w hich used parameters of g eneralized Gaussian density functio ns fitting subband histograms as features and part it ioned histog ram f itt ing metho d. Shape features w ere ex t racted using a new -proposed deformable model method aiming at cloud object s, called circle unit method. A fter image distance or similar ity r ules w ere given for each metho d, experiment s w ere made. Result s indicate that the applicat ion of CBIR in satellite cloud imag e retrieval is ef fective, and that special ly desig ned shape metho d achieves much bet ter perfo rmance than common methods due to the use o f satellite cloud imag e domain know ledg e, though at the same t ime mo re t ime-consuming.