基于分块颜色直方图和GWLBP的图像检索算法

为了提高多特征融合图像检索的效果,本文提出了一种基于分块颜色直方图和GWLBP的图像检索算法。算法采用K-means均值聚类对RGB颜色空间进行颜色聚类,再将4×4均匀分块图像分成9个子块,提取每个子块的颜色体积直方图,并赋予不同权值计算颜色特征;利用Gabor滤波器组对输入图像进行不同分辨率和方向滤波,然后将不同方向上局部滤波器输出结果与全局滤波器输出结果的平均值进行比较,并进行二值化,据此提出3种不同的GWLBP算子来提取纹理特征。最后对图像的颜色和纹理特征高斯归一化,采用加权平均来融合颜色和纹理的特征距离。通过实验仿真可知,与其他3种算法相比,本算法对正常和有旋转倾向的图像都有较高的查全率和查准率。

Image retrieval algorithm based on block color histogram and GWLBP

In order to improve the effect of multi-feature fusion image retrieval, this paper proposed a kind of image retrieval algorithm based on block color histogram and GWLBP. By means of K-means mean value clustering, the algorithm implemented color clustering on RGB color space; then, it divided 4×4 uniformed block image into 9 sub-blocks to extract color volume histogram of each sub-block and endow different weights so as to calculate color features. Based on Gabor filter bank, the smoothing in various resolutions as well as directions were conducted on input image; later, the comparison of mean values between local filter output results and overall filter output results in different directions was carried out with threshold. On this basis, three kinds of GWLBP operators were proposed to extract textural features. Finally, Gaussian normalization was implemented on image color as well as textural features; weighted average was applied to integrate feature distance between color and texture. The experimental simulation indicated that, compared with other three algorithms, this algorithm carries higher recall ratio and precision in normal images and those with rotation inclination.