Automatic seismic image segmentation by introducing a novel strategy in histogram of oriented gradients

Automatic seismic image segmentation and further geological interpretation, requires accurate detection of the target's boundary through applying image analysis techniques. Various types of image segmentation and analysis methods are available for this purpose, using the texture attributes, the machine learning methods or the computer vision tools. The histogram of oriented gradients (HOG) as an advanced image analysis tool is frequently used in the pattern recognition investigations. Here a novel strategy in the extraction of the texture attributes based on using the concept of the HOG is introduced for automatic seismic image segmentation and properties extraction. The proposed strategy consists of extracting the HOG features, deriving statistical parameters related to the texture attributes and separating the target by integrating the selected images. To increase efficiency of the proposed strategy, a new parameter as the hybrid texture attribute is also introduced. The proposed method is applied on two geometrically simple and complex synthetic models and two field seismic data examples, containing saltbody and mud diapirs. The saltbody in the first selected field data example demonstrate a chaotic pattern with the unsharp boundary. The mud diapirs in the second field data example illustrate the mild chaotic pattern with interdigitated boundary and thin marginal flows. The result of applying the presented strategy on the both seismic images, compared to the result of the conventional interpretation methods, texture of attributes (TOG) and other textural attributes, revealed that it can be considered as an alternative for conventional image interpretation. However, it should be noted that at this stage, the presented method can be used for separating geological objects with any shape, different seismic patterns and sufficient contrast with surrounding media. However, it is not yet developed for fault detection and horizon auto tracking.

Automated summary

A novel strategy based on the concept of HOG for texture attribute extraction is proposed for automatic seismic image segmentation and properties extraction. The method is applied on both synthetic models and field seismic data examples, showing promising results as an alternative for conventional image interpretation.