In order to solve the problems of phase unwrapping low precision induced by real phase loss, error transmission, and excessive smoothing while we use (0-1) mask and least square iterative method to deal with phase unwrapping of local high density residual point wrapped phase, a new phase unwrapping method is presented. By using (2k+1)×(2k+1) sub pixel discrete Gaussian convolution kernel to mask the high density residual point region, the true phase information of the region is effectively preserved. The weight of the four directions least square iterative method is set by adjusting degree-maximum phase gradient deviation. Therefore, the over-smoothing of the least square iterative method is improved and the error propagation in the mask area is restrained. The accuracy of the phase unwrapping of the high density residual point area is effectively improved. The experimental results show that this method can solve the phase unwrapping problem of high density residual point in the local region effectively and the original phase of high density residual point can be restored quickly and effectively. Compared with the traditional least square iterative method, the average error of the high residual point region of the proposed method is only 10% comparing to least square iterative method under the same iteration times. This method is more suitable for accurate phase unwrapping of high density residual point region.