基于多尺度分解的图像递进检索

提出了一种基于小波多尺度分解的图像递进检索匹配方法.首先用小波变换对图像进行多尺度分解,提取其各子带的前三阶矩(标准差、偏斜度)和最低频段(LL子带)的小波变换值作为特征值.然后在检索匹配时,用各尺度的标准差和偏斜度进行递进筛选,再用LL子带的HSV颜色直方图计算图像之间的相似性距离,进一步对检索结果求精.与传统的直接直方图方法比较,结果表明,运用该方法,不仅计算量大大降低,明显加快检索速度,而且检索精度也明显提高.同时,由于小波变换已经成为新一代图像压缩标准的核心,该方法可以直接在压缩域进行特征提取和检索.

Image Progressive Retrieval Based on Multi-scale Decompositions

A method is proposed for matching the image progressive retrieval based on multi-scale decompositions.The image is decomposed first with wavelet transform to extract the moment (standard and skewness) of each sub-band and the lowest frequency (LL sub-band) as characteristic value of wavelet transform.When matched retrieval is conducted,progressive filtration is processed by using multi-scale moment,and then HSV bar graph of the LL sub-band is used to calculate the similarity distance between the images for further refining the retrieval result.Comparing with the traditional direct bar graph,the result shows that the calculation is greatly reduced,the retrieval is speeded up obviously and its accuracy is improved by using the method.Because the wavelet transform is the kernel of image compression standard of the new generation,the method can be used for direct extraction and retrieval in compression domain.