Speaker: Fabrice Taty-Moukati

Date: Thursday 6th of January 2022, 1:15 pm.


Building the data energy term of marked point processes is of an utmost importance. The main reason is that characterizing objects on images, in the framework of marked point processes, requires building a probability density, which is often composed of two terms. The first one is the interaction energy characterizing interactions between objects and the second one is the data energy which contains the locations as well as sufficient information of objects. The goal of our work is to detect and characterize seismic faults in the context of image analysis to deals with fault uncertainty problems. Seismic images themselves are subject to uncertainties, mainly due to processing steps as well as the limited seismic bandwidth. For these reasons, we cannot rely on a single interpretation to perform analyses of any field or reservoir, even if obtained with neural networks, since we know that this interpretation can comprise ambiguities. We also know that geological faults, as one may take a closer look at them, are not linear but damage areas composed of several small cracks. Based on the facts set out above, we adopt the foregoing approach based of segment processes to characterize and detect seismic faults. We do that since a seismic fault network can be seen as a realization of a marked point process. One of our goals is to increase the resolution of seismic faults that we seen in seismic images. For doing so, we start with fault probabilities provided by FaultNet3D [Wu et al.,2019] to better constrain the data energy term in accepting or rejecting segments.