The edge detection enhancement on satellite image using bilateral filter

Satellite imagery is the image taken from satellites from outer space. It captures the appearance of the earth’s surface through remote sensing. Usually, satellite image’s object is hardly detected because many objects are covered with cloud shadows in the sky. Edge detection has an important role in image analysis. Edge detection aims to extract the boundary of an object contained in the image. Gaussian Filter is usually used on edge detection to smooth the image and reduce noise. However, in satellite image, we can improve it by using another filter. In this research, we compare gaussian filter with bilateral filter and analyze those two methods. We also using several operators to see the optimized process. After the experiment, by using the comparison parameter such as largest PSNR value, and the smallest MSE value, we can conclude that Bilateral Filter with Canny Operator is the most optimized edge detection for Satellite Imagery.


Introduction
Satellite imagery is very useful in various sectors such as geography, mining, and energy, meteorology, weather, forestry, etc. Satellite imagery is difficult to detect because many of the satellite images are covered with cloud shadows in the sky [1]. There are many methods for edge detection. Some methods for edge detection are using Roberts Operator, Canny Operator, and Frei Chen Operator. These methods have different characteristics. Edge detection becomes the main approach in image recognition and analysis can be based on the edge of the image containing useful information [2]. Image enhancement using the Gaussian Filter is commonly used for edge detection to reduce image noise. Gaussian filters are applied to smooth the image and reduce the noise, but will also smooth the edges [3]. Bilateral filter is a technique to smooth images while preserving the edges [4]. This technique is very similar to Gaussian filtering [5]. Bilateral filter is a non-iterative, local and simple nonlinear technique by combining gray degrees based on geometric proximity and photometric similarities [6]. Therefore, this paper proposes an optimized algorithm for satellite image edge detection based on image enhancement using Bilateral Filter by preserving edges. The purpose of this research is to analyze traditional edge detection of Gaussian Filter on satellite images and improve edge detection with Bilateral Filters, then the comparative analysis is conducted in this research to see the optimized edge detection.
In image processing, filtering is one of the bases used in various tasks such as interpolation, resampling, and noise removal. The filter used depends on the nature of the task performed by the filter and the type of data to be used for the operation [7]. Filtering is one tool used in many fields for various applications, especially in this section for image enhancement.

Gaussian Filter
Gaussian Filter is used in the field of image analysis especially for smoothing process, blurring, eliminating detail, and noise removal [8]. In Gaussian Filter, the intensity value of each pixel is replaced by the average of the weighted value for each of its neighboring pixels and pixels itself. The number of neighbors involved depends on the filter being designed. For digital image processing, the zero mean Gaussian of two variables is expressed by the equation: Discrete estimates of this continuous function are determined using two parameters [9] (1) the desired size of the kernel (as N x N filter mask); (2) the value of , the standard deviation of Gaussian function.

Bilateral Filter
Bilateral filters are one of the noise reduction methods that have two kernel filters: spatial kernel and range kernel [5]. The spatial kernel aims to smooth the image based on the Gaussian function [10]. Spatial Kernel is the distance between the pixels of the image (Euclidean distance), and range kernel is the similarity of the intensity between the two pixels in the image. Spatial kernel ( ) is the realization of spatial proximity measurement is by the equation While range kernel ( ) means the weighting of pixels in accordance with the size of the difference in intensity of the pixel with the intensity at the pixel which is the center of analysis on the image. Calculation of range kernel in each pixel is shown in the equation.
Bilateral Filtering ( ( )) computation as follows: Where I is a grey level image, N(p) is a neighborhood of p, Ipis the intensity value at pixel location p, Wp is the normalization factor which ensures pixel weights sum to 1

Edge Detection
Edge is the essential characteristic of an image [4]. The edges of an image contain crucial information from the image Edge detection is a significant step in image processing. Image edges can represent the objects contained in the image, shape, size, and sometimes also information about the texture. Edge detection is used to extract the essential features [14].

Roberts Operator
The first operator used for edge detection is the Roberts Operator [11]. It was given by Robert Cross in 1963. Roberts operator is a first-order operator, which uses a partial differential operator to find the 1234567890''""  [2]. Roberts operator is a simple edge detection technique and has a fast computing rate. The image to be detected by the edge is combined with the mask of the operator and the gradient magnitude to be calculated directly. The Roberts operator uses the following 2x2 masks:

Canny Operator
Canny is one of the modern edge detection algorithms. In 1986 John Canny proposed three criteria to be the basis of filter mining to optimize edge detection in noise-sensitive images. The first step in implementing Canny is to smooth the image [2]. Then find the image gradient in both vertical and horizontal directions [12]. Canny uses a Gaussian filter before applying the mask. Gaussian filters can reduce the noise as much as possible [11]. Canny edge detectors are not sensitive to noise causing Canny can detect weak edges [2].

Frei Chen Operator
Frei Chen is a unique operator [12]. Frei Chen mask contains the base vector. This is implemented in the 3x3 image area represented by the number of nine Frei Chen masks [13]. Here we described the frei chen operator used in this research :

Slope Magnitude Method
Slope magnitude can use the following equation where Gx and Gy is the image gradient in the horizontal and vertical direction [12]. Calculation of Gradient Magnitude (significant slope of the line) is denoted by Equation 6 below: Where: = gradient magnitude = gradient to point x = gradient to point y

Mean Square Error (MSE)
MSE is the mean of the square of the error value between the original image and the image of the image processing. The greater the value of MSE, the worse the quality of the resulting image.

1234567890''""
10th Conversely, the smaller the value of MSE, the better the quality of the resulting image. Mean Square Error (MSE) can be formulated as follows [11]: Where: = the length of the image in pixels = image width in pixels ( , ) = pixel value after edge detection ( , ) = pixel value after edge detection

Peak Signal to Noise Ratio (PSNR)
Peak Signal to Noise Ratio (PSNR) is a calculation that determines the value of an image produced. The value of PSNR is determined by the size or magnitude of the MSE value that occurs in the image. The greater the value of PSNR, the better the resulting image. Conversely, the smaller the value of PSNR, the resulting image gets worse. PSNR can be calculated by the formula [14]:

Results and Discussion
The dataset used in this research are satellite images which has been resized into 300 x 300 pixels resolution. We retrieve the data from USGS (United States Geological Survey). In this research we use lakes and seas object from the image. Figure 1 shows the comparative result between gaussian filter and bilateral filter  The result shows edge detection on Roberts, and Frei Chen produces a smooth edge of the line and seen several lines cut off. While Canny produces more edge. Clouds on satellite imagery are also considered as edges.The higher the value of given to the Gaussian Filter will remove the edge of the image.
In table 2 we only use different values for each image and keep the value. The image on the edge detection of Roberts and Frei Chen has increased even the edge of the line on Frei Chen is thicker than using Gaussian Filter. Edge detection with Bilateral Filter can retain the edges although on Canny is not very visible.
We also experiment with increasing value of on the Bilateral Filter. The result shows that if the value of higher than before, the noise on the image becomes faded because controls the power of the noise filter on the image. As a result, the detected image loses many edges. While at , the higher the value of , then the object in the image faded. But it can still keep the edge so edge information can still be seen.