Combining noise-adjusted principal components transform and median filter techniques for denoising modis temporal signatures

Consistent multi-temporal images are necessary for accurate landscape change detection and temporal signatures analysis. Orbital images have a difficulty to maintain a temporal information precision due to several interferences that generate missing data. In this paper is developed a program in C++ language for denoising MODIS temporal signatures considering two-phase scheme for removing impulse and white noise. In the first phase, the median filter is used to identify impulse noise. In the second phase, the Noise-Adjusted Principal Components (NAPC) transform is applied to eliminate white noise. Because they are two complementary methods, there is high performance in removing noise. The restored NDVI (Normalized Difference Vegetation Index) signatures showed a significant improvement providing a time series dataset that can be used to identify and classify the vegetation physiognomic types.