The inversion of Albedo, of AMTIS image based on the assumption of student-t observation error distribution

In paper "Validation of observation error distribution and parameter distribution of linear kernel-driven BRDF model", we concluded that: when freedom degree is big enough, the goodness of fit of student-t distribution and normal distribution is equivalent, but student-t distribution has better robustness; from statistics value, the priori distribution of parameters is neither normal distribution nor t distribution, but student-t distribution has better goodness of fit and better robustness than normal distribution. [1] In paper "The EM algorithm and it's application in quantitative remote sensing inversion", corresponding to student-t error distribution we put forward EM algorithm and validate it's precision and robustness in theory using the 73 dataset which can be used to produce prior knowledge. As we all know, the 73 dataset have different precision with air-borne and space-borne images and there are different in application and theory. In this paper, we will use four algorithms to inverse the albedo of a 400*400 AMTIS image and compare their precision and robustness.