|
|
|
Poster Session 2
|
Automatic Cloud Cover Assessment for SPOT Images
Method
The method if this study consists of two major steps : (1) threshold segmentation and (2) post processing .
- Threshold segmentation
the segmentation includes the design of a spectral threshold. The key 9ot select the threshold is based on the fact that the cloud pixels normally have higher spectra responses than other features in an image. in this study , the empirical tests suggest that the threshold can be selected as the image's mean pluses on e standard deviation. Therefore as long as the pixel has the value greater than the threshold, The pixel is considered as being the cloud pixel. Moreover, the empirical tests indicate that this threshold condition has to applied to SPOT's all three spectral bands in order to correctly locate the cloud pixels. The aim of this step is designed to rap only cloud pixels.
- The post-processing
the step is designed to ding out more cloud pixels that are excluded form the initial segmentation stage. The method used here is the region growing algorithm. The algorithm uses the cloud pixels located in the first stage as the seed pixels to grow around the neighboring area. Thus, based on a user-defined criteria, pixels with similar spectral values that are adjacent the seed pixels become the cloud pixels. Since the region growing process may include some highly reflective landscapes, most of them are the bare land of the river bank, it is necessary; to eliminate these non-cloud pixels . because the shapes of most of the sub-tropical clouds are different from the rivers which mostly present as the linear shapes in image, a linear shape template is used to move around the areas that have grown in the previous step. The pixels that have grown in a linear shape will then be eliminated.
|
|
|
|
|
|
|
