综合颜色和轮廓曲线特征的图像检索方法研究

摘要

For the overall contourlet transform is redundant and the wavelet transform cannot efficiently take advantage of the fact that the edges usually found in natural images are smooth curves.In this paper,a new image coding scheme is pre-sented.The wavelet transform is applied to original image,and the directional filter banks(DFB) are employed to middle and high frequency subbands;The visual gains are achieved by exploiting HVS characteristics to weigh wavelet coefficients according to their perceptual importance; the new extended spatial orientation tree are constructed.The lower frequency exploit a new partitioning structure based on the importance of coefficients.The experiment results show that the modified image compression scheme provides higher perceptual quality and higher PSNR than SPIHT and WBCT,especially at low bit rates.%传统的基于内容图像检索(CBIR)及跟踪算法主要利用图像的颜色、纹理等特征进行相似性比较,但大量的实验和应用也表明利用颜色和纹理进行图像相似性比较在空间结构和对象形状上难以精确控制,致使图像检索经常出现一些不可预料的结果.为了提高图像在形状、颜色及纹理上的检索精度,提出了一种综合颜色和图像轮廓曲线特征的检索方法.该方法分割图像并提取图像中感兴趣对象的轮廓,对提取的轮廓进行仿射变换及最小值化处理,经处理后的轮廓带有边缘的完整信息,具有几何不变性;利用聚类的颜色信息,提取主聚类的直方图,所提取的直方图不仅包含了主聚类的颜色信息也包含了该聚类的空间位置信息.利用检索对象与被检索对象的颜色距离直方图及轮廓曲线距离偏差的加权平均度量检索及被检索对象的相似性.实验结果表明,针对基于感兴趣对象的图像检索问题,给出了一种具有高度检索精度的算法.