Object recognition of high resolution remote sensing image based on PSWT

Abstract

High resolution image provides an important new data source for object recognition. A method based on pyramid-structured wavelet transform (PSWT) for object recognition on high-resolution remote sensing image is put forward. First, the model image and the search image are decomposed to pyramid-structure by wavelet transform. A novel match metric is computed using direction vectors constructed by the high-frequency information of wavelet coefficients. This metric is robust to occlusion, clutter, illumination changes. And the matching can be implemented until the highest level of decomposing to track said instances of the model in the lowest one. The tests confirm that the proposed algorithm is efficient and reliable.