更新时间：04-14 （白色球鞋）提供原创文章
摘要:现在做的局部特征大多数是基于角点和尺度空间划分出局部的区域，提取能够仿射不变的局部信息特征，如SIFT，然后进行匹配，这种算法计算量非常大，要存储的特征向量的长度也是非常长，通常都在上千。对于内部细节不是很丰富的样本，这种方法不经济。本文主要研究内容检索的目标图像，尤其是对分块检索方法进行了研究。基于分块图像特征的图像检索方法将目标图像划分为多个分块，每个分块图像特征反映了图像的局部特征，而多个子块图像特征的结合又能够对区域整体形状进行描述，由于四叉树分块方法的欠缺，极坐标下分块方法相对准确。首先对目标图像进行归一化处理，以目标图像的最小外接圆作为目标区域，然后在极坐标下分块，对于分块后的子图像，我们通过实验充分比较这些子块图像的不同局部特征函数来测试比较。
关键词：图像检索 ；最小外接圆；子块图像特征；极坐标下分块
ABSTRACT:Now most of the local features based on the angle point and scale space are carved out of the local area which can extract the local affine invariant information. Such as SIFT, and then proceed to match, but this algorithm is very large, the length of the stored feature vectors is very long, usually in the thousands, this method is not economical. In this paper, we studies mainly about the content retrieval target image, especially the block retrieval method. The image retrieval method based on image features of the block is that the target image is divided into multiple blocks. Each sub-block image features reflect the local features of the image. And the combination of multiple sub-block image features can describe the shape of the region as a whole. Due to the lack of block of the quad tree, the block method in polar coordinates is relatively accurate. First of all the target image is normalized, then see the smallest circum-circle of the target image as a target area, and divide image block into blocks in polar coordinates. For sub-image after block, through the experiment we can fully compare the accuracy of these different local characteristics of the sub-block image function.
Keywords: image retrieval; smallest circum-circle; block of image features; polar coordinates block