Statistical threshold for real time pattern matching using projection kernels

This paper is concerned with real time pattern matching using projection kernels. We derive an analytical threshold based on statistical properties of random noise and characteristics of the projection kernels. The proposed threshold decision scheme provides a mean to perform a utomatic pattern matching without human intervention. Based on the required successful rate, the analytical threshold can reliably reject mismatch and keep the target pattern irrespective of the assumption of noise model. Experimental results show that the proposed threshold follows the ground truth threshold tightly, and the false rejection rate is less than 1% even the image is very noisy.