Gabor Filters for Segmentation of Salt Structures

The suitability of Gabor filters for segmenting a salt structure present in a North Sea seismic section is investigated. A filter bank of 24 Gabor filters is designed and applied to the seismic data. Principal component analysis is applied to the resulting filter response images to extract a low number of relevant attributes. These attributes are used to train a statistical classifier to estimate the probability that each pixel is salt. The probability is then used as input to a segmentation algorithm that finds a continuous, smooth border between the salt and the surroundings. The resulting border corresponds well with a manually interpreted contour provided by an experienced interpreter.