基于Vague集模糊量化的图像检索算法

An image retrieval algorithm based on vague set fuzzy quantization and generalized histogram

为了提高图像的检索效果,提出了基于Vague集模糊量化的图像检索算法。该算法首先在色调-饱和度-亮度(hue saturation value,HSV)空间中划分模糊区域,并建立每个区域的梯形隶属度函数;其次,通过梯形隶属度函数得到任意颜色在模糊区域的Vague隶属度,进而将该Vague隶属度转化为包含中立度信息的Fuzzy值,并对该Fuzzy值加权得到新的量化值,从而得到颜色直方图;最后,将原图平滑处理,并与原图直方图结合得到广义直方图,通过欧氏距离进行相似匹配得到检索结果。实验采用Corel1000图像库,选取花作为检索的目标图像,结果显示,非均匀量化算法,模糊量化算法和改进算法的查准率分别为62.5%,72.5%,75%;查全率分别为25%,29%,30%。采用3种算法分别检索主目标明确和主目标不明确的图片,结果显示,改进算法与非均匀量化算法,模糊量化算法相比,检索效果较好。

In order to improve the image retrieval performance,an image retrieval algorithm based on Vague set Fuzzy quantification is proposed.In this algorithm,firstly,the Fuzzy region is divided into HSV(hue saturation value)space and the trapezoidal membership function of each region is established.Secondly,the membership of the Vague of any colour in the Fuzzy region is obtained by the trapezoidal membership function,and then the membership of the Vague is transformed into the Fuzzy.The Fuzzy value of the information is obtained and the new Fuzzy value is weighted to obtain a new colour histogram.Finally,the original image is smoothed and combined with the histogram of the original image to obtain a generalized histogram,which is matched by the Euclidean distance to get the search results.The Corel1000 image database is used in the experimentand the flower is selected as the retrieval target image.Simulation results show that the precision of the non-uniform quantization algorithm,the Fuzzy quantization algorithm and the improved algorithm are 62.5%,72.5% and 75%respectively,and the recall rates are 25 %,29%and 30%correspondingly.These results show that the improved algorithm is better than the nonuniform and Fuzzy quantification algorithms in retrieving the main target and therefore it is the ambiguous target algorithm.