Normalization of Remote Sensing Imagery for Automatic Information Extraction
For the time being, Remote Sensing automatized techniques are conventionally designed to be used exclusively on data captured by aparticular sensor system. This convention was only adopted after evidence suggested that, in the field, algorithms that yield great resultson data from one specific satellite or sensor, tend to underachieve on data from similar sensors. With this effect in mind, we will refer to remote sensing imagery as heterogeneous.There have been attempts to compensate every effect on the data and obtain the underlying physical property that carries the information, the ground reflectance. Because of their improvement of the informative value of each image, some of them have even been standardized as common preprocessing methods. However, these techniques generally require further knowledge on certain atmospheric properties at the time the data was captured. This information is generally not available and has to be estimated or guessed by experts, avery time consuming, inaccurate and expensive task. Moreover, even if the results do improve in each of the treated images, a significant decrease of their heterogeneity is not achieved. There have been more automatized proposals to treat the data in the literature, which have been broadly named RRN (Relative Radiometric Normalization) algorithms. These consider the problem of heterogeneity itself and use properties strictly related to the statistics of remote sensing imagery to solve it. In this master thesis, an automatic algorithm to reduce heterogeneity in generic imagery is designed, characterized and evaluated through crossed classification results on remote sensing imagery.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)