Comparison of thresholding techniques for TV-to-IR image registration

The most widely used method for accomplishing digital image registration in real-time is correlation. Correlation algorithms can be easily implemented in real-time using simple digital hardware if the gradient images are quantized to bi-level or tri-level. Several optimal and suboptimal techniques of quantizing gradient images to bi-level or tri-level are considered in this paper. Simulation results indicate that a thresholding technique based on measurable statistics such as mean and standard deviation of gradient values yields better correlation results than a thresholding algorithm which yields a prespecified percentage of each level in the quantized gradient image.