A Study of Joint Histogram Estimation Methods to Register Multi-Sensor Remote Sensing Images Using Mutual Information

Registration is the basic image processing operation in a variety of tasks such as multi-source classification, image fusion and change detection. Automatic intensity based registration techniques are gaining importance. In this paper, we investigate an intensity based technique that utilizes mutual information as the similarity measure. We apply this technique to perform multi-sensor image registration. The performance of a number of joint histogram estimation methods for the determination of mutual information has been evaluated using a measure called registration consistency. These methods include partial volume interpolation, cubic convolution interpolation, linear interpolation, and nearest neighborhood interpolation. The results show that partial volume interpolation produces the most reliable registration consistency. Nearest neighbor interpolation outperforms linear and cubic convolution interpolation.