ABSTRACT
Images rendered from measurements made by wireline microresistivity imaging tools include longitudinal gaps whenever the well circumference exceeds the total width of the pad-mounted electrode arrays. The fraction of an image containing null data depends on tool design, and it is commonly approximately 30% for 261 mm (8.5 in.)-bit size wells increasing to approximately 50% in 311 mm (12.25 in.) wells. We use cues from the measured parts to infer information missing from the gaps; a method has been developed that simulates the process by decomposing the measured parts into their morphological components using sparse representations of multiscale multiorientation transforms, then recomposing the full-bore image assuming it to be efficiently represented by the transform’s elemental bases. The approach was evaluated using real data sets with a variety of geologic features, including full-coverage images from small diameter wells artificially obscured to simulate images from larger diameter wells. For borehole images dominated by curvilinear features, reconstructions from artificially obscured images were visually indistinguishable from the original unobscured images for a broad range of coverage loss and for all apparent dip angles below near-vertical, regardless of degree of parallelism (or lack thereof). Successful reconstruction of near-vertical features (including those with complex boundaries such as breakouts) was more dependent on coverage loss, but in these cases, the results were consistent with judgments made by interpreters. Therefore, we found that inpainting provides a consistent starting point for reproducible quantitative geologic analysis, and it is an enabler for automated feature recognition.
