Fundamental Performance Limits in Image Fusion
The task of image fusion is fundamental in image processing. It is a necessary preprocessing step for many modern image processing and computer vision tasks, and many algorithms have been developed to address the image fusion problem. Often, the performances of these techniques have been presented using a variety of relative or experimental measures comparing different estimators, leaving open the critical question of overall optimality. In this paper, we address this open question by modeling the image fusion problem through the use of Markov Random Fields (MRF). A Bayesian formulation is presented in which a likelihood term is defined using an observation model based on coefficient and geometric fields. These coefficients, which represent the individual contributions of the source images, are modeled as the prior information by means of Markov Random Fields. This probabilistic approach allows the identification of optimal estimators by minimizing an energy function in terms of both fields, making the fusion of the images possible. The fundamental performance limits for the problem of image fusion are derived from the Cramer-Rao inequality. Quantitative evaluation and visual effects show that our model achieves both better image fusion performance and lower computational cost than the traditional pixel based algorithms.
Erwin Gilmore Mohamed Chouikha
Howard University Washington, DC/United States Howard University Washington, DC/United States
国际会议
The 31st International Congress on Imaging Science(第31届国际影像科学大会 ICIS2010)
北京
英文
285-288
2010-05-12（万方平台首次上网日期，不代表论文的发表时间）