抽象的な

An Accurate Subpixel Shift Registration in Noisy Image Using a Kernel Regression Method

Hossam Eldeen M. Shamardan

In this paper, a new accurate subpixel registration for pure shift estimation is proposed. The noise effect, which disturbs the quality of registration process , is taken into account. The kernel regression method which represents the field of nonparametric statistics is used as a tool for the estimat ion process due to its powerful capabilit ies in the field of both denoising and interpolation. The kernel regression depends on studying a local region intensities distribution and gradients. By applying gradient descent method, the global translation parameters can be estimated. Experimental results show that our proposed method can estimate the translation parameters accurately. Furthermore, our method performs well in noisy images.