Seismic Profile Image Texture Segmentation Method

Seismic section image has distinct texture feature, and the different region of texturerepresent different geological structure. The break of texture direction and structure meansthe break of geological structure. This information is important to search for oil and gas. Soin this thesis we propose a method used in seismic section segmentation, which has notonly theoretic meanings but also potential applied value. The research works conducted inthis paper are summarized as follows:(1) The description and extraction of image feature is important to texture imagesegmentation and recognition. Because the texture feature provides the essential structureinformation in different regions of a texture image. So this paper introduces the method ofthe extraction of texture feature firstly. On the base of fully understanding their advantageand disadvantage, we choose the fractal model to extract image feature considering theseismic section image.(2) The accurately estimate to Fractal dimension (FD) using fractal model is the keyto extract texture feature.We particularly discussed the general methods of estimatingfractal dimension, and then, we introduce an improved method of estimating fractaldimension. Furthermore, we present a new approach of estimating fractal dimension,which related to mathematical morphology. After the comparison with the previousmethods, the result of experiment show that the fractal dimension based on mathematicalmorphology can be effective as texture image features.(3) For the purpose of reducing border estimating error in segmentation, we use the8-neighborhood edge-preserving noise smoothing quadrant filter (EPNSQ), instead of thetraditional 4-neighborhood technique, to smooth each of the fractal dimension featuresbefore practical segmentation. A Self-Organizing Feature Map neural network algorithm isused for segmentation. We perform experiments with three different methods on extractingand segmentation feature of real seismic section image. The results of the experimentsprove that the methods we used are effective.