基于多样字典理论与多尺度距离度量的彩色图像检索

The color image retrieval algorithm based on multiple dictionary theory and multi-scale distance measure

针对在图像检索中因色彩因素导致的相关算法正确率低、稳定性差等问题,提出了一种多样字典理论与多尺度距离度量的彩色图像检索算法.首先,对输入图像进行量化,将其转换为一维字符串形式;其次,采用多样字典统计对图像视觉模式编码,并计算编码后的图像特征值;最后,给出多尺度距离的相似度量准则,并根据该准则对查询图像与数据库图像的特征值进行处理,寻找与其匹配的特征图像.实验结果表明:本文所提出的算法在查准率与查全率上要优于当前流行的检索方法,其对彩色图像的检索精度和稳定性也有了明显提高,因此具有较好的应用价值.

A new color image retrieval algorithm based on the theory of multiple dictionaries and multi-scale distance measure was proposed to improve the low accuracy and instability which are the problems caused by color factor.Firstly,quantifying the input image and converting it to a string form;secondly,encoding the image visual pattern by calculating the multi-dictionaries,and then evaluating encoded image feature value.Finally,this paper presents a similarity metric for multi-scale distance.According to the criterion,the characteristic values of the query image and the database image are processed,and then a matching image is found to complete the search task.Experimental results shows that this algorithm proposed in this paper was superior to the current popular methods in precision and recall level and the accuracy and stability of the color image retrieval has been significantly improved.Therefore,the algorithm has good application value.