结合MSA与方向编码的快速景象匹配

Fast scene matching combining MSA and direction code

针对MSA(Multi-Scale Autoconvolution)算法对光照变化敏感且在进行目标搜索时计算量大的缺点,提出了一种结合MSA和方向编码的快速景象匹配算法。首先,求待匹配图和模板的方向编码图,以消除光照变化的影响;其次,对方向编码图进行分块处理,分别求取各子块的MSA值,在每个搜索位置,只更新MSA值矩阵的一行(列),这样使计算量减少到不足原MSA算法的1/8。用cameraman图像和实际卫星拍摄图像进行实验,结果表明,改进后的算法不仅对仿射变换、轻微透视投影变换、噪声及遮挡具有鲁棒性,同时可以抗光照变化的影响,且搜索速度更快,在实际拍摄的图像中能对目标进行准确识别和定位。

In order to solve the problems that Multi-scale Autoconvolution（MSA） algorithm is sensitive to an illumination change and has a large computation load,a fast scene matching algorithm combining the MSA with a direction code is presented.Firstly,the direction code images of an image to be matched and a template are computed respectively to eliminate the influence of illumination.Then the code images are blocked and the MSA value of each sub-block is calculated.So,only one row or one column of MSA value matrix needs to be updated at every searching position.In this way,the computation load has been decreased greatly,and it is less than 1/8 of that of the original MSA algorithm.Experiments are carried out with a cameramen image and real scenes.The results show that the proposed method is robust to affine transform,light projective transform,noise,occlusion and illumination change,meanwhile it has a faster searching speed and can accurately recognize and locate objects in real scenes.