GISdevelopment.net ---> AARS ---> ACRS 1992 ---> Digital Image Processing

Cloud cover and could shadow removing based on 2-dimensional histogram

F. Cheevasuvit, K. Dejhan, T. Tanapanpanich
Faculty of Engineering, King Mongkut's Institute of Technology Ladkrabang
Ladkrabang, Bangkok 10520, Thailand

D. Lisawadiratanakul
Faculty of Engineering, Kasem Bundit College
Patanakarn, Bangkok 10250, Thailand.


Abstract
This proposes a method for cloud cover and cloud shadow removing by using 2-dimensional histogram. Two satellite images of visible band, which acquired at different time and different cloud distributions, have been used. The first axis of histogram is formed by a satellite image which is preferred to remove its cloud and shadow. While the second axis is formed by the difference image of these two satellite images. The cloud and shadow pixels in preferred image can be detected by applying two sets of thresholds to the 2-dimensional histogram. Therefore the cloud cover and cloud shadow pixels in the preferred image will be substituted by the correspondent pixels from the other image.

Introduction
The removing cloud cover from satellite imagery is very useful for assisting image interpretation. The average cloud coverage for the entire world is about 40% …. Especially for Thailand, the cloudy weather is appeared persistently. The interpretation form the cloudy image can not effectively achieved. Many researches have been proposed for cloud removal. Most of them used a single threshold to detect the cloud in the meteorological satellite image 12,3,41. The threshold has been applied to the histogram of the data for segmenting the scene. Since the clouds boundaries are diffused, hence it is difficult to detect all cloud types by, using a single threshold. A dual thresholding method have been developed by[5]. This method can detect the cloud in a picture provided the cloud does not exist permanently in the same place. The first threshold is a common threshold which found by examing the histogram plot for detecting the cloud pixel in the interested image. While the second threshold is obtain from two registered image. This threshold corresponds to a minimum possible difference value between cloud and cloud-free pixels in the same position of two register images. Therefore the cloud can be removed from the first image by replacing cloud-free data of the second image. Since the second threshold is applied to the histogram of absolute difference between two registered images. These two threshold are simultaneously applied to two histograms which not convenient and also the absolute difference value in the second histogram can not give any information by itself. This method can only remove the cloud but not its shadow. To improve the mentioned defects, we propose a 2-dimensional histogram method for detecting and removing cloud and its shadow therefore, two group of thresholds have been used for detecting cloud and its shadow.

Methodology
Two visible satellite images acquired at different times with different cloud distribution as show in Fig. 1 (a) and 1 (b) have been used to creat a 2-dimensional histogram. The first axis of this 2-dimensional histogram is formed by the distribution of pixel intensity of the first image which preferred to remove the cloud and its shadow. In this image the cloud pixels have high intensities or they will be sited in the bright portion of the histogram. The cloud shadow pixels have low intensities, so they will be sited in the dark portion of this axis. The second axis of the 2-dimensional histogram is formed by the difference of these two satellite images in the different image, the position of cloud pixels will give a high positive values. While the cloud shadows will give a middle negative values as shown in Fig 2. Therefore two groups of thresholds will be selected. The first group is used fro detecting cloud pixels and the second group is used for cloud shadow pixels.

The method can be described as follows:
  • Rectify and register two satellite images by using correlation technique.
  • Brightness matching of images by using their mean brightness and standard deviation from the overlap regions. This can be don by where x is old brightness of a pixel and y is its new value of the second registered satellite image. m… and s… are the first registered satellite image value of mean brightness and standard deviation and m1 and s1 are the