Abstract: The default surface generation is one of the core issues in 3D seismic data interpretation.The traditional method is that the geologists create contour curves in planes according to the slice and coherence information，and conjecture the three-dimensional shape of the fault.This method is a time consuming step of the 3D data interpretation，and result in ambiguity.The matching between frames of image sequences technique is widely used in motion tracking，motion compensation，multi-source image fusion，video compression，multi-frames super-resolution image and other fields.This paper presents a new method based on motion estimation to track the feature points and generate the fault surface.This method is like human interpretation processing，and fully considered the relationship between the sequences.And the experimental results prove the validity of this method.

Key words: seismic image sequence, motion estimation, subpixel matching, down-sampling in frequency domain, H∞ suboptimal filter

摘要： 在三维地震数据解释中，断层面的识别及建模是核心问题之一。传统的地质断层曲面的生成依靠地质研究人员根据地质剖面和相干切面的信息，用人工方法识别断层轮廓线，再推测断层面的三维形态。这种方法在三维数据解释中最费时，许多地质解释数据也会有人为的二义性。图像序列的帧间匹配及运动估计技术已经广泛运用于运动跟踪、运动补偿、多源图像融合、视频压缩、多帧图像超分辨等领域。提出了一种基于运动估计的断层面生成方法，跟踪地震图像中断层的特征点并生成断层曲面。这种方法完全符合人工解释理解模型，并充分考虑了图像序列的帧间关联。实验结果表明，该方法是有效的。

关键词: 地震图像序列, 运动估计, 亚像素精度匹配, 频域下采样, H∞次优滤波器