- In many image processing applications, the estimation of the covariance matrix is considered an essential step. Estimating the covariance matrix has a great influence on the success or failure of a given algorithm. Usually the covariance matrix is estimated by the sampled covariance matrix of the whole data. The problem with doing so is that anomalies that exist in the data might distort the covariance matrix. This paper presents an approach for covariance matrix estimation that is less prone to anomalies and improves the detection rate. Results on simulations and real life images are presented.