Abstract:
Thin cloud is one type of the most common noises in optical remote sensing images. It will severely affect the recognition precision of images. As thin cloud mainly exists in low frequency domain, the algorithm with adaptive thresholds of wavelet transform is presented to remove thin cloud in the single band image. The thresholds between thin cloud and land surface objects in the wavelet coefficients can be adaptively searched with general crossing validation (GCV) and genetic algorithm (GA). Then thin cloud is removed by shrinking wavelet coefficients. Results indicate that thin cloud can be effectively removed with the proposed algorithm while land surface objects are maintained, and blurry details of land surface objects become clear. With the highest entropy, the cloud-removed image from the algorithm shows better effect than those from homomorphic wavelet-based filter and homomorphic filter. The different scale thresholds of wavelet coefficients between thin cloud and land surface objects can be picked out with GA and GCV automatically.