Image retrieval is an important research topic in pattern recognition and computer vision. We propose a novel content-based image retrieval method. At the image description stage, our method first modifies the traditional micro-structure descriptor (MSD) to capture the direct relationship between the shape and texture features and that between the color and texture features. Our method then extracts the uniform local binary patterns (LBP) histogram of the image to capture the color difference information. At the image comparison stage, our method first compares the descriptors of the images to compute their similarities. The similarity between each pair of images is then updated by also considering the similarities to comparable images within the dataset. Accordingly, our method obtains the final similarities of the images. Experimental results demonstrate the effectiveness of our method for image retrieval.

图像检索是模式识别和计算机视觉中的重要研究课题。我们提出了一种新颖的基于内容的图像检索方法。在图像描述阶段，我们的方法首先修改了传统的微结构描述符（MSD），以捕获形状和纹理特征之间以及颜色和纹理特征之间的直接关系。然后，我们的方法提取图像的统一局部二进制图案（LBP）直方图以捕获色差信息。在图像比较阶段，我们的方法首先比较图像的描述符以计算它们的相似度。然后，还通过考虑与数据集中可比较图像的相似性来更新每对图像之间的相似性。因此，我们的方法获得了图像的最终相似度。