Abstract: The classic similarity function adopted in general bilateral filter is yet subject to image noise,it fails frequently to reflect the grey similarity between two pixels.This paper proposes a new estimate of the similarity between two pixels based on the ordered statistics of their grey values.Due to the correlation among the adjacent image pixels,the grey values of all pixels in the 1-ring neighborhood of the concerned pixel can be regarded as its multiple observations,which is referred to as its n-ordered statistics value.We re-define the similarity between two pixels in the bilateral filter according to the Euclidean distance of their n-ordered statistics values.This novel estimate of similarity is based on the grey distribution within the 1-ring neighborhood of the two pixels thus is able to restrain the noise influence.Combining the new grey similarity estimate with the metric of spatial proximity suggests a robust bilateral filter.Experiments results demonstrate a better performance of the proposed filter than that of current approaches.