Next, draw in the stick like wooden framing which will support the person who will be sitting in this beach chair.
Development of Shoreline Extraction Method Based on Spatial Pattern Analysis of Satellite SAR Images
The extensive monitoring of shorelines is becoming important for investigating the impact of coastal erosion. Satellite synthetic aperture radar (SAR) images can cover wide areas independently of weather or time. The recent development of high-resolution satellite SAR images has made observations more detailed. Shoreline extraction using high-resolution images, however, is challenging because of the influence of speckle, crest lines, patterns in sandy beaches, etc. We develop a shoreline extraction method based on the spatial pattern analysis of satellite SAR images. The proposed method consists of image decomposition, smoothing, sea and land area segmentation, and shoreline refinement. The image decomposition step, in which the image is decomposed into its texture and outline components, is based on morphological component analysis. In the image decomposition step, a learning process involving spatial patterns is introduced. The outline images are smoothed using a non-local means filter, and then the images are segmented into sea and land areas using the graph cuts’ technique. The boundary between these two areas can be regarded as the shoreline. Finally, the snakes algorithm is applied to refine the position accuracy. The proposed method is applied to the satellite SAR images of coasts in Japan. The method can successfully extract the shorelines. Through experiments, the performance of the proposed method is confirmed.
Keywords:
Introduction
In recent years, coastal erosion has led to the decline of sandy beaches. The decline of sandy beaches makes seawalls and roads along the coast more vulnerable to damage from waves. Coastal erosion is caused by various factors, such as decreases in the supply of sediment coming from rivers due to the construction of dams and changes in the flow of drifting sand caused by the construction of ports. Coastal erosion manifests as changes in the shoreline, and, therefore, wide area monitoring of shorelines has become more important for identifying the factors causing coastal erosion and for establishing effective countermeasures. Monitoring requires continuous observation in wide areas. Satellite imaging is suitable for monitoring owing to its low cost for taking images and its wide coverage area. Optical satellite images, however, cannot be used to observe the shoreline when they are taken at night or when the area is covered with clouds. On the other hand, images acquired by synthetic aperture radars (SARs) are more suitable for monitoring owing to the following characteristics: they can operate during the day and night, their observations are not influenced by the weather, and they can accomplish regular monitoring according to their orbit.
Research on automatic shoreline extraction from SAR images has been carried out. Lee and Jurkevich presented an early study on automatic shoreline extraction from satellite SAR images [1]. Because their research aimed to obtain the direction and position of the coastline for autonomous navigation and the determination of the geographical position of ships, the monitoring of shorelines was not taken into account for automatic extraction. Their method was sufficient for obtaining the direction and position of the coastline, but thin inlets could not be successfully extracted because of certain problems, such as speckle and low SAR image resolution. Methods for automatic shoreline extraction have been proposed since that research was published. These methods can be divided into two types of approaches: (1) extracting the shoreline as edges [2,3,4,5] and (2) classifying the sea and land areas and then extracting the shoreline as the boundaries between them [6,7,8].
In an approach of the first type, Mason and Davenport [2] used low-resolution images by taking the average pixel value in a 4 × 4 region to eliminate speckle and reduce the necessary calculation time. Edge extraction was then applied to these low-resolution images [9], and the results were superimposed on the images with their original resolution. Finally, the snakes algorithm, proposed by Kass et al. [10], was applied to determine the final shoreline. However, there were limitations on the settings for edge determination and for the snakes algorithm due to speckle, and extraction errors occurred in many parts. Liu and Jezek [3] proposed eliminating the speckle first and then applying edge emphasis and the Canny filter [11] as an edge extraction method. Results were acquired through threshold processing. This method was applied to SAR images of Antarctica as an experiment, but it was difficult to extract shorelines if there were obstacles, such as icebergs, in the coastal area. Wang and Allen [4] and Asaka et al. [5] used observation data from the PALSAR (ALOS “Daichi”), which is a satellite SAR that was launched by Japan. Wang and Allen [4] used the images obtained by outputting the maximum and the minimum pixel values within a specific filter size and taking the difference. They exploited the fact that this difference increases in the vicinity of the shoreline. The influence of speckle is suppressed by outputting the maximum and the minimum values within the filter size. Asaka et al. [5] selected the edges using a one-dimensional Laplacian of Gaussian (LoG) filter to obtain sea/shoreline/land training data for extracting shorelines. In these experiments, it was possible to extract the shoreline in general. However, if the shoreline was unclear because of speckle, there were some cases in which the shoreline could not be accurately extracted.
In the other type of approaches, the sea and land areas are first classified and then the boundary between them is extracted as the shoreline. Zhang et al. [6] performed segmentation exploiting the fact that the distribution of pixel values in sea areas is more homogeneous than in land areas after performing smoothing for speckle removal. Using this method, relatively homogeneous land areas and non-homogeneous sea areas may be erroneously classified, but they can be removed based on their small size. Nunziata et al. [7] used multi-polarized SAR images. By exploiting the fact that the correlation between VV and VH (the first capital letter indicates the angle of transmission while the second letter denotes the angle of reception with H and V, horizontal and vertical angles, respectively) is more remarkable than single-polarized scattering intensity images, sea and land areas were segmented. Vandebroek et al. [8] monitored the changes in shorelines to evaluate the results of a beach nourishment project (Sand Motor) on the west coast of the Netherlands. In approximately 55% of the images, they were able to effectively extract the shoreline. Ultimately, they concluded that coastal dynamics could be evaluated only for changes in the order of tens of meters.
The spatial resolution of the first satellite SAR launched for the observation of Earth’s oceans (called Seasat) was approximately 15 to 30 m, while that of satellite SARs launched in recent years is approximately 3 to 5 m. For shoreline monitoring, the accuracy required depends on the spatial resolution of the satellite SAR images used. The erroneous extraction of patterns (such as wave crest lines and sandy beaches) cannot be ignored when working with SAR images of higher resolution. SAR images with coarse resolution or small influence of waves were used in most previous research. Therefore, there are few studies mentioning the erroneous extraction due to wave crest lines or sandy beach patterns. Vandebroek et al. [8] extracted shorelines using TerraSAR-X images with a resolution of 3 m for shoreline extraction. The influence of the waves (in the sea area) and the brightness of sandy beaches (in the land area) were pointed out as factors of failure in extraction. Accordingly, when conducting wide-area monitoring with a unified method, countermeasures against wave crest lines and sandy beach patterns are essential.
Smoothing filters, such as the Gaussian filter, are effective for removing speckle. However, parts of the wave crest lines and sandy beaches are often extracted incorrectly as edges. Figure 1 shows an example that the ordinary method has a limitation against the effect of textures such as wave crest lines and sandy beaches. Wave crest lines and sandy beaches seem to have different spatial patterns from that of speckle. It is difficult to remove them in the same way as when removing speckle. On the other hand, they seem to repeat similar patterns. Speckle is considered to be a section where pixel values change locally, while the shoreline is considered to be a part where large changes of pixel values are linearly connected. By separating speckle, wave crest lines and sandy beach patterns from SAR images and using only components related to the shoreline, shorelines can be extracted with high accuracy. The proposed method in this paper deals with image decomposition to separate such texture patterns from original SAR images.
The term “shoreline” in this paper is used according to the Vandebroek et al. [8]. It is defined as the instantaneous intersection of water and land at the time of the SAR observations.
The SAR data used in this research are described in Section 2.1. The principle of the proposed method and our method of accuracy verification are explained in Section 2.2. Our results and discussion are presented in Section 3. Finally, Section 4 concludes this paper.
Materials and Methods
2.1. Input Data
We used the SAR data observed by the L-band synthetic aperture radar PALSAR-2, which is installed in a satellite called ALOS-2, launched by the Japan Aerospace Exploration Agency (JAXA) in 2014. Most of the SAR scenes were obtained with the “Ultra-fine” observation mode (single look) that has a spatial resolution of 3 m. Full polarization scenes were obtained under the observation mode of “High-sensitive”, with a spatial resolution of approximately 5 m in the range direction and 4 m in the azimuth direction. The observation area of each single SAR scene under the “Ultra-fine” mode is 50 km by 70 km (range by azimuth), while the “High-sensitive” mode has a slightly narrower observation extent in the range direction. Because the higher resolution was required in this study, only HH-polarized waves (“Ultra-fine” mode) were used, and multi-polarized observation data were not used in combination. PALSAR-2 data have a processing level indicating how it was processed from the raw data. In this research, we used processing levels 1.1 and 1.5. Processing level 1.1 consists of complex-valued data obtained after range compression and azimuth compression and includes phase information in addition to intensity. The horizontal direction of these SAR images is the slant range direction. On the other hand, processing level 1.5 contains only intensity information, but the horizontal direction in this case is the ground range direction (after map projection).
The coasts of interest are shown in Table 1 and Figure 2. In order to measure the actual position of the shoreline, on-site observations were conducted for each coast. We used a GARMIN GPSmap 60CSx (Handy GPS) (Olathe, KS, USA) to measure the shoreline. The accuracy of the GPS is ±3 m, which is almost equal to the resolution of SARs. On-site observations were made close to the planned time for satellite observation so that the position of the shoreline observed from the satellite coincided with our on-site observations. Whenever the time of an on-site observation differed from that of the satellite observation, that on-site observation was made in a time such that the tide level during the satellite observation coincided with the tide measured on-site, either several days before and after. In Sakata, the tide level during the on-site observations was not consistent with the tide level during the satellite observation, and thus the SAR images of Sakata were excluded from our quantitative accuracy verification. In this research, we verified the accuracy of the proposed method by comparing the shoreline extracted using it with the GPS data of the on-site observations. The method for accuracy verification is described in Section 2.2.5. The SAR images were trimmed to 256 × 256 pixels.
2.2. Methods
The proposed method consists of four steps. As mentioned in Section 1, we focused on spatial patterns in SAR images. The shoreline is considered as the part of the image where large changes of pixel values are linearly connected, and speckle is considered to be the parts where the pixel values change locally. Wave crest lines and sandy beaches seem to have similar repeating patterns. By using a sparse-modeling technique [13], images including only smooth components, such as the shoreline (outline), and images including speckle and spatial patterns (texture) are decomposed. Only the outline component is used in subsequent steps.
The outline image is smoothed using the non-local means filter. Then, the image is segmented into sea areas and land areas (hereinafter called “sea–land binarization”), and the boundary is extracted as the shoreline. For the sea–land binarization process, the graph cuts technique is applied. The graph cuts technique can reflect the distribution of pixel values and their difference with surrounding pixel values. Finally, the extracted shoreline is refined using the active contour method.
In the following sections, the principle of each method will be explained step by step.
2.2.1. Image Decomposition Based on Spatial Patterns
First, image decomposition is applied based on the differences in spatial patterns, such as the shoreline, speckle, wave crest lines, and sandy beaches in the SAR images. Because the objective of this research is the extraction of shorelines, it is necessary to decompose the source data into images containing only the components related to smooth changes, such as shorelines, and images containing speckle and other spatial patterns. We assumed that both types of images are sparsely generated from different models [14]. This sparsity means that most of the elements of the feature vectors representing the image are equal to zero. Such techniques based on sparsity are called sparse modeling techniques. Morphological component analysis (MCA) [15,16,17] is one of such sparse modeling techniques, and it is suitable for our image decomposition problem. MCA requires a dictionary, which is a matrix of parameters representing spatial patterns of image intensity. The dictionary should be determined beforehand. In our research, a dictionary was learned using small patches for each image. Then, each SAR image was decomposed into an image including only components related to smooth changes, such as the shoreline (outline), and an image including speckle and spatial patterns (texture). The learning process and the image decomposition were accomplished as follows.
Let y 0 ∈ ℝ N be an image of size N × N pixels and composed of the outline component y c and the texture component y t (i.e., y 0 = y c + y t ). Let us consider a small patch p ∈ ℝ n of size n × n pixels ( n ≪ N ) , where p is a column vector in which pixel values are arranged in a row. These patches have a dictionary A ∈ ℝ n × m ( n < m ) . Moreover, each patch p has a corresponding sparse representation q , which satisfies
How to Draw a Shore Line
In this post I will show how to draw a shoreline. Main element in such compositions is the wave of water reaching the shore. This has to be portrayed correctly to give the overall impression of shore. This is shown step by step below.
Start by laying down the curvature of shore as this will help you put down the curvature of wave next. In this example I used rocks to create a rocky shoreline. Draw wave with a broken wavy line representing wave front and indicate flow of water using broken lines as shown below. Click to see details.
Prev
1 of 6
Next
Add Other elements:
Once you learn to draw wave, you can use different settings for your shore. Here I start by adding small stones and using dots and ticks to give indication of sand/shore.
A backdrop of pine trees and hills is added to complete the setting. Drawing these different elements is discussed in my workbooks and Free tutorials.
Adjust Tone:
As there is no erasure in pen and ink drawing, it is always good to start by using less tone and then progressively adding more to your liking. Here I felt that there wasn’t enough contrast between stones and surroundings and so I added more tone to the stones to create contrast and interest. This completes this drawing.
This completes this tutorial. Use the template for your attempts. This is not an easy drawing and getting correct feel for water wave will require multiple attempts and practice. Enjoy the creative process of drawing and feel free to reach out to me for help and suggestions.
I imagine my beach somewhere in the Bahamas so here draw in the stem or trunk for the palm tree.
You will now draw in the palms to complete your palm tree like so, then add small detailing to the leaves which are notches.
Add more leaves to the tree like so, then proceed to step thirteen.
Here is how your beach scene looks when you are all done. Now you can color in your drawing.
Comments (0)
More From Dawn
Related
Details
Description: The weather is starting to really warm up which means city and state beaches will start piling in the people. Today I will show you “how to draw a beach scene” and I know I’ve done beach tuts in the past but they weren’t too colorful and didn’t really have the “beach” setting. In this lesson I have a lunge chair, umbrella and beach ball. Even though I’d like to have my own quiet spot on the beach all to myself, I know in reality that isn’t possible. For some reason I notice more and more people just go to the beach to lay in the sun, read a book, and to build sand castles. Although I do see some folks swimming in the water, you don’t see as many that should. Anyways, this is a colorful, simple lesson that anybody can tackle. Have fun with it, and happy summer folks!
Login
Not a member?
Leaving Dragoart for External Site
We temporarily stopped you from leaving Dragoart so you could confirm
The site above is not affiliated with Dragoart in any way. Please continue at your own risk. Don’t enter any passwords or personal information from a site claiming to be Dragoart or its affiliates.
