International Journal of Computer Theory and Engineering

Abstract

Interpolation methods play an important role in many fields such as industrial, geological and military fields for prediction. However, it is quite difficult to predict the unknown information only by some sparse hard data in the process of simulation based on current popular interpolation methods. Accuracy of simulated images can be improved by using soft data and hard data. Multiple-point geostatistics (MPS) originates from geostatistical fields and allows extracting multiple-point structures from training images, after that MPS can copy these structures to the regions to be predicted. To simulate or predict information accurately, an interpolation method using soft data and hard data in MPS is proposed. Dimension reduction is made by filters to reduce the CPU time and memory demand. All similar training patterns fall into a cell in the filter score space, which is created by filters. Finally, a training pattern is randomly drawn from a cell, and then is pasted back onto the unknown region to be predicted. The variogram curves of the simulated images are compared, showing that the structural characteristics of the image simulated by using both soft data and hard data are most similar to those of the training image.