A target sample x(i,j) of a two-dimensional array of reconstructed samples is filtered based on values of samples in a neighboring region of the target sample to produce a two-dimensional array of modified reconstructed samples, according to the equation: y(i,j)=round(x(i,j)+g(Σm,n∈Ra(m,n)ƒ(x(i,j)−b(m,n)x(m,n)))), where y(i,j) is a modified target sample value, R is the neighboring region of the target sample, a(m,n) and b(m,n) are real-valued coefficients, round(x) is a function that maps the value x to an integer value in the range [0,2B−1], B is the number of bits representing each sample of the two-dimensional array of modified reconstructed samples, ƒ(x) and g(x) are functions, wherein (a) ƒ(x) is a non-linear function, or (b) g(x) is a non-linear function and both a width and a height of R is more than one sample.