Abstract
The measurement of shoreline is a key factor in coastal-zone construction. The traditional ground survey is time and cost consuming. In this study, we propose a automatic scheme to detect shoreline for SPOT multi-spectral images. This scheme contains two stages. Firstly, we convert the color model of image representation from RGB to ISH and then use the ISODATA to classify saturation information to obtain the areas that contain the potential tideland information. Secondly, we use the hue information to refine the tideland areas and retrieve the shoreline. The test result shows that the proposed scheme is useful and practical.

Introduction
The estimation of shoreline is very important for the coastal construction. To construct a facility near coastal area we need to realize the short-term(several weeks) and long-term(several years) variation of shoreline about area of tideland. Traditionally, using ground survey to measure the position of shoreline is labor intensive and time consuming. Because of the vast ground coverage and highly temporal resolution of SPOT satellite, it is both time and cost effective to use SPOT multi-spectral images to detect shoreline, even though the positioning accuracy is poorer than ground survey due to its the 20m resolution.

It is very difficult to automatically identify the tideland from SPOT images if we use image classification techniques. The main reason is that they normally need some sample information as training data for supervised classification or as references to assign class for unsupervised classification. In this study, we develop an automated scheme to delineate tideland from multi-spectral SPOT image and then obtain shoreline.

Method
SPOT multi-spectral images contain 3 spectral bands. These bands cover green(0.50~0.59mm, band 1), red(0.61~0.68mm, band 2) and near infrared(0.79~0.89m m, band 3). In general, there are many color models to represent a color image. Normally, RGB color model is used to represent SPOT images. For example, we use the gray scales of red, green, and blue color to display band 3, band 2 and band 1 of a multi-spectral SPOT image, respectively. In this study, we use another color model, ISH[ 1], to represent SPOT satellite images. The formula to convert RGB to ISH is as follows:

Intensity(I):

Saturation(S):

If M = m, S = 0         (2)

Hue(H):

If M = m, H = 0                     (8)

If R = M, H = 60(2+b-g)         (9)

If G = M, H = 60(4+r-b)         (10)

If B = M, H = 60(6+g-r)         (11)
where the range of R, G and B is from 0 to 1.0. M is the maximum value in R, G and B. m is the minimum value in R, G and B.

By inspecting the saturation(S) of SPOT images, it can be found out that there are value gaps between sea water, tideland and land areas. The relation of saturation value between these classes can be expressed as:
S_sea>S_tideland>S_{land area}        (12)
Then, we can use ISODATA unsupervised classification method to classify the saturation image into 3 clusters. Accordingly, the cluster that has the mean in the middle can be identified as the potential tideland areas. The cluster has maximum mean is categorized as the sea areas. The last one with a minimum mean is classified as the land areas.

Since some land areas may be misclassified as the tideland, the potential tideland obviously need further refinement. In this study, the hue(H) image is used to implement the refinement, because the following characteristic can be found:
H_{sea & tideland} > H_{land area}
In the following stage we classify hue image into 2 clusters using ISODATA. Following the classification, we can use the classified land area from hue image as a mask to wipe off the misclassied area from potential tideland area as a refinement procedure. Finally, we clump refined tideland area into patches and select the patches adjacent to sea area as tideland. Then, we can extract the edge[ 2] of tideland areas as the shoreline.