2. Texture Content
The textural information is assumed to be contained in the overall, or average, spatial relationship among gray levels for a particular image. Although no formal definition of texture exists, intuitively this descriptor provides the measures of properties, such as smoothness, coarseness and regularity. The three-principle approaches use to describe the texture context are statistical, structural and spectral. The statistical approaches use to be descriptor of spatial relationships and yield characterizations of textures as smooth, coarse, grainy, and so on. This
spatial relation is considered to be the covariance of pixel values as a function of distance and direction between pixels. Such information can be extracted from an image using gray-tone spatial-dependence matrices or co-occurrence matrices (Wisetphanichkij, 1999). Let Z be a random variable denoting discrete image intensity, P be a position operator and let A be a k*k matrix whose element aij is the number of times that points with gray level zi occur (in the position specified by P, relative to points with gray level zj, with 1 =i, j =k. The co-occurrence matrix (A) can be constructed as follow;
1. The numbers of different pixel values (zi) are determined.
2. These pixel values are ranked (k) smallest to largest.
3. The digital image is scanned in the direction noted (P-operator) to determine the frequency with which one of these pixel values follows another.
4. Each entry in matrix (A) is divided by n, the number of pixels in the image satisfying P, let this resultant matrix be called co-occurrence matrix (C).
To analyze a co-occurrence matrix (C) in order to categorize the texture, the statistical parameters as a set of descriptors are computed as follows.
1. Maximum probability
maxij(Cij)
2. Second-order inverse element difference moment.
3. First-order inverse element difference moment,
4. Entropy
5. Uniformity
3. Image Fusion
The image fusion can be divided into two classes: spatial domain method and spectral domain method. The last method is used in most application, such as color space transformation. In this paper, the IHS (Intensity-Hue-Saturation) model will be used as a color space and the image fusion is done as follow:
1. The RGB color space of OPS image is transformed to the IHS model [3]:
Equation - 6
Equation - 7
and
Equation - 8
2. The different gray value of pixel in the black-white of two SAR images (g1 and g2) are added into the OPS image intensity:
Equation - 9
The last term of the above equation is the different of before and during flood. The flood area will be emphasized and non-flood area will be depressed. Adding this term to intensity component in IHS mode means transferring of flood area data to OPS image.
3. The IHS model is inversely transformed to the RGB space and ready to classify in the further.
4. Neural Network Classification
In this paper, the multi-layer perceptron (MLP) neural network based on back propagation (BP) algorithm is used as classifier, which consists of set of nodes arranged in multiple layers with connection only between node in adjacent layer by weights. The input informations are presented at input layer as the input vector. The output vector is the processed information, that are retrieved at the output layer. The hidden layers stand between these two-layers. A schematic of a three-layer MLP model is shown in Fig 4 and used in this paper.
In this work the nonlinear function, Sigmoid function given in Eq.(12), is used to determine the output state:
Equation 12
Figure 4 The 3 - layer (MLP) model of neural network.
