Mostly, shape from focus (SFF) methods utilize initial depth estimate to obtain 3D shape of an object. However, accuracy of these methods is limited due to erroneous initial focus and depth measurements. In this paper, we introduce a Gaussian process regression based approach, which estimates 3D shape of the object from the noisy initial depth values and focus measurements. Initial depth is estimated by applying a conventional focus measure. Eigenvalues from 3D neighborhood around the initial depth are computed to form the input feature vectors. A latent function is developed through Gaussian process regression to estimate accurate depth through these features. The proposed approach takes advantages of the multivariate statistical features and covariance function. The proposed method is tested by using image sequences of various objects. Experimental results demonstrate the efficacy of the proposed scheme.