The Pixel Relativity (PR) measurement of SAR image, which is the key of the despeckling techniques based on weighted average, is researched in three aspects. Firstly, the rationality of ratio PR model is expounded, and two new ratio PR models, which are the LOG-domain Gaussian model and the pixel similarity probability model, are proposed. Meanwhile, the Probability Density Function (PDF) of SAR image and the PDF of the ratio between pixels are transformed into ratio PR models. Then, in order to evaluate the four ratio PR models, the weighted maximum likelihood filters are designed using the PR. Finally, a novel method, performed by calibrating the maximum location of the PR model, is introduced to improve the radiation preservation of those models whose maximum do not locate at 1. The effectiveness of the two proposed PR models and the approach to calibrate the maximum location of the PR model, are indicated by the theoretical analysis and experimental comparison.