Motivated by the representation of biometric and multimedia objects, we consider the problem of hiding noisy point-sets using a secure sketch. A point-set X consists of s points from a d-dimensional discrete domain [0, N -1]d . Under permissible noises, for every point x1, .., xd X, each xi may be perturbed by a value of at most . In addition, at most t points in X may be replaced by other points in [0, N - 1]d . Given an original X, we want to compute a secure sketch P. A known method constructs the sketch by adding a set of random points R, and the description of (X R) serves as part of the sketch. However, the dependencies among the random points are difficult to analyze, and there is no known non-trivial bound on the entropy loss. In this paper, we first give a general method to generate R and show that the entropy loss