Seismic Data Resizing Using Seam Carving

Seismic data interpretation can be accelerated by reducing the size of the seismic images. Cropping, decimation, and resizing (decimation after averaging), are some well-known techniques to accomplish these goals. Rather than regularly removing certain rows and columns (decimation), or predefined groups of rows and columns (cropping) from the input image, seam carving (SC) discards the least informative regions of the input image while preserving its content. SC is a computer vision algorithm originally devised to resize natural images that, to our best knowledge, had not been applied to resize geophysical data previously. We propose to adopt SC to deliver smaller, yet representative, seismic images that contain the most relevant structures and textures of the original data with amplitude preservation. These reduced images can be used by seismic interpreters to cope with day-to-day Industry demands and deadlines since they can analyze the results of filters and seismic attributes with less computational resources and in shorter time frames. Essentially, SC relies on an energy operator and dynamic optimization to find the optimal pixel trajectories to carve from the input image iteratively until the desired image size is attained. We modify the original SC algorithm by using an energy operator based on the well-known Sobel magnitude that grants high reduction rates in both dimensions with minimum artifacts. We illustrate SC by means of a toy example and an image from the Moon’s surface. Then, the proposed method is applied to resize field seismic data that comprise two seismic sections and one seismic amplitude time-slice. The results demonstrate that SC, contrarily to decimation or conventional resizing, provides meaningful reduced seismic images that preserve the main features of the original data.