Analysis of the Matching Suitability of Remote Sensing Radar Images of Different Sensors Based on Sketch Token

Abstract:

Focusing on the extraction and description of common information between heterologous images, this paper proposed an algorithm for matching suitability analysis between heterologous images based on Sketch Token middle level features. Supervised learning has been used to acquire priori knowledge of homogeneous information between heterologous images, and utilize this knowledge to train the Sketch Token classifier, which is designed as a descriptor of homogeneous information to extract features on reference images. The statistical of extracted features is applied for training support vector machine classifier, which is used to analysis the matching suitability of reference images. The algorithm is validated by the satellite SAR(synthetic aperture radar) images which have been used as reference images and aerial RAR(real aperture radar) images which have been used as real images. The test result demonstrate the effectiveness of the algorithm.