基于方向可控金字塔二值图像投影的图像检索

Image Retrieval Based on Steerable Pyramid Binary Image Projection

针对目前基于内容的图像检索方法中存在的图像纹理信息表征不全面、运算量大、检索效果不理想等问题,提出了一种新的图像检索方法。方法首先对灰度图像进行方向归一化,然后进行方向可控金字塔分解,得到多尺度多方向的子带图像,通过自适应阈值T进行子带图像二值化并分别在行方向和列方向上进行投影得到图像的纹理特征。在特征匹配上,采用向量相交匹配方法。实验结果表明,上述方法运算速度快,检索效果和抗干扰性好,可以用于图像检索。

To solve the existing problems such as incomprehensive in texture expression, huge computation and ineffectiveness in content based image retrieval, this paper proposed a new image retrieval method. In this method, first a gray-scale image was normalized in orientation. Then, it used steerable pyramid to decompose the image into sub-band images in different scales and orientations. These sub-band images were converted to binary images by the adaptive threshold value T. The binary images were then summed in row and column to get texture features. Vector intersection method was defined in the process of matching. Experiments show that the proposed method is fast. It gets a good retrieval result and is robust to common interferences.