17-10-2016, 03:20 PM
1459474724-paperami.docx (Size: 100.01 KB / Downloads: 5)
Abstract
The mainstay of the project is finding the oil slick features in VIIRS nighttime images .satellite images are captured and processed to find the exact areas of oil like and oil spill regions in oceans, where the problem arises particularly for nighttime images .It is based on statistical information and clustering of those areas based on the density of spills, as they spread densely in water. It is demonstrated to be able to effectively remove the noise in those images and identify line features. The Visible Infrared Imager Radiometer Suite (VIIRS) has similar orbital characteristics and a wider swath than Moderate Resolution Imaging Spectroradiometer (MODIS) and thus is also suitable for oil spill detection under sun glint and cloud-free conditions. Indeed, VIIRS imagery collected at night may be also used to capture oil slick features under moon glint. This letter proposes a practical method to extract oil slick features in a semiautomatic fashion. however, the requirement of human intervention to determine optimal parameters points to the need for improved automation in future works.
Introduction
OIL spills can cause serious damage to marine and coastal ecosystems. Natural oil leakages account for a large percentage of crude oil releasing into the ecosystem. Hence, ecological imbalance can occur at a geological time scale .Satellite images of reflected sunlight have been used to detect and monitor oil spills in oceans. A major obstacle in the application of oil slick feature extraction from optical remote sensing image is image noise and other natural features. Results in either a low signal-to-noise ratio or other image features such as clouds or cloud shadows .The problem is particularly severe for nighttime images captured by the Visible Infrared Imager Radiometer Suite (VIIRS) .Remote sensing techniques, with the development of state of-the-art sensors and algorithms, serve as effective means to monitor and assess oil pollution in oceans and to guide coastal resource management. . An efficient algorithm to reduce noise contamination is required and to extract the oil slick features, particularly from low-SNR images. Most satellite observations of oil spills in oceans rely on synthetic aperture radar (SAR) data [6] because of their high spatial resolution, transparency to most clouds, and day–night operability .Where as the noise strongly influences the feature extraction results of SAR images, geometric filters and edge detection algorithms are often required .This system captures the input image by VIIRS is filtered using Gaussian filter to remove the noise and to smoothen it .Later image is enhanced by using Histogram equalization. Then K-means clustering algorithm is used to cluster the oil spill area based on their densities of the oil spills and identifies even the minute area of spills in water.
Preprocessing
It is the lowest level of abstraction. The aim of pre-processing is to improve the image data that suppresses unwanted distortions and enhances some image features more important for further processing .In this processing we have to apply the speckle noise to the image. Then the noisy image is given to the Gaussian filter noise will be removed .It is further used for the calculation of a new pixel brightness, geometric transformations, pixel brightness transformations .This is because the noise pixels have similar pixel values with oil slicks and thus will interfere with the feature extraction. The RGB image is converted to gray scale during preprocessing by using “rgb2gray” convertor.
Guassian Filter:
Guassian Filter is used to blur images and remove noise and defines a probability distribution for noise or data. It smoothens noise efficiently.The Gaussian function is used in numerous research areas. Gaussian kernel coefficients are sampled from the 2D Gaussian function,
G(x,y)=1/2(pi)(sigma)2*e(-x2+y2)/2(sigma)2
where σ is the standard deviation of the distribution.Gaussian function is never equal to zero. It is a symmetric function. The distribution is assumed to have a mean of 0. Graphically, the familiar bell shaped Gaussian distribution is,
The Gaussian filter works by using the 2D distribution function with the image.This requires an infinitely large convolution kernel, as the Gaussian distribution is non-zero everywhere. Fortunately the distribution has approached very close to zero at about three standard deviations from the mean. This means we can normally limit the kernel size to contain only values within three standard deviations of the mean. In this case, even after the parameters were optimized, the WMM filter did not lead to better results than a Gaussian filter .This might be due to the imperfect assumption in modeling noise distributions. Because oil slick features often have relatively narrow widths (one to several pixels) particularly when they come from natural seeps, noise filtering should both preserve the narrow features and avoid breaking coherent line features. A Gaussian filtering window can be applied to achieve this. Slick features inevitably become exaggerated after spatial smoothing. Thus, a threshold filter and low-intensity (referring to a relatively small smoothing window) spatial smoothing are applied to allow the discovery of slick-like features.
Histogram Equalization:
Histogram equalization is a technique for adjusting image intensities to enhance contrast and noise removal image is enhanced by using histogram equalization. This enhanced is process under the thresholding. After removing the isolated noise pixels, histogram equalization, with the appropriate cut-down value, is applied to the AOI. .The equalization is based on the area’s statistical range after which dark slick features are more distinct from the background . The statistical range is selected based on the AOI’s histogram. Typically, 0.5 × maxvl is the cutoff value for bright pixels, and minvl is the cutoff value for darker pixels. Next, a threshold filter is used to filter all bright pixels while retaining the darker pixels. The threshold value is set as
thr = A × stdev ,
where stdev represents the standard deviation of the AOI, and
A is a scaling factor ranging from 0.2 to 0.8 for different images.
Because no direct correlation between this threshold and either the mean or the stdev value of the AOI has been found, a derived scaling factor relationship cannot be applied to every image. This is typical for image segmentation. The image segmentation performance of the threshold-filterbased approach was compared with other well known segmentation methods, such as the level set method, the mean shift, the Markov random field, and fuzzy c-means methods. Those methods yield less satisfactory performance than our proposed method when applied to VIIRS nighttime imagery.