25-08-2017, 09:32 PM
1456490096-CHAPTER1.docx (Size: 94.17 KB / Downloads: 6)
INTRODUCTION
With the technological developments in aerospace an increasing number of Earth observation commercial satellites with high-resolution sensors have been launched such as Quick Bird (QB) IKONOS and WorldView-2 (WV-2). The images obtained from these satellites have very high spatial resolution (VHSR) usually ranged from 0.5 to 4 m. At this resolution details such as buildings and other infrastructures are easily visible. Therefore these VHSR images have opened a new era for remote sensing applications such as object detection classification object mapping and change detection. In particular VHSR images have attracted much attention from researchers studying urban areas due to theexistence of relatively small features such as roads buildings bridges and trees. Inevitably tall standing objects whichmainly are buildingsamong these small features cast longshadows in most of the captured VHSR images. These shadows may be utilized as a valuable cue for inferring 3-D scene information based on their position andshape for example for building detection and building height estimation. On the other hand the shadows cause partial or total loss of radiometric information in the affected areasand consequently they make tasks like image interpretationobject detection and recognition and change detection more difficult or even impossible. In this project we focus on the second aspect of shadows i.e. to attenuate the problems causedby the loss of radiometric information in shadowed areas bycompensating or reconstructing them. Generally two steps areinvolved in this procedure: First shadow detection and secondly shadowreconstruction or compensation.
Regarding shadow detection in VHSR images two mainapproaches are reported in the previous literature namely themodel-based and the property-based. The former requires priorknowledge of scene or sensorsincluding but not limited todistribution of scene radiance and acquisition parameters likesun azimuth sensor/camera localization date and the timeof day of acquisition. Based on the prior information themodel-based approaches obtain good performance in detecting particular type of objects like buildings and vehicles.
These approaches are not general enough to deal withthe great diversity of geometric structures which usually existing VHSR satellite images of urban areas. The property basedapproaches make use of certain shadow properties in images such as brightness spectral characteristics and geometry. Because of their simplicity both in principle and implementation the property-based approaches have been widely used in literature; they generally include four categories:
1) Thresholding-based
2) Color-transformation-based
3) Region-growing-based
4) Classification-based
In the thresholding-based methods the shadow and non- shadow pixels are determined according to a predefined threshold level which usually can be set according to the bimodal distribution of image histogram such as the method. Inthe color-transformation-based methods the red–green–blue (RGB) color image is first transformed to a 3-D space suchas hue–intensity–saturation hue–saturation–value (HSV) andYCbCrmodels; then a new image is derived according tospecific spectral properties of shadows in new space suchas shadow areas having lower intensity higher hue values and higher saturation.Finally shadows are detected bythresholding the derived new image. The proposed shadowdetection method in first transformed the RGB image intoHSV space and then derived a normalized saturationvaluedifference index (NSVDI) to identify shadows via thresholding. Several photometric invariant color models for shadow detection were compared. In the region growing basedmethods the seed points are first selected and then each of thepixels is assigned to a segment according to their distance fromthose regions to which they could potentially be assigned. The classification methods can also be employed for shadow detectionbecause of the commonly possessed properties in shadowedareas such as their lower intensity. Recently the author’s inhave proposed to utilize the support vector machine (SVM) classification method for shadow detection in which a binaryclassification procedure was implemented in a supervised manner to derive a shadow-versus-non-shadow mask.
In order to reconstruct the detected shadowed areas threealgorithms were introduced namely the gamma correction method the linear-correlation method and the histogram matching method. The gamma correction method consideredthe shadow as a multiplicative source that corrupts the brightness of the underlying pixels and then built the relationshipbetween shadow and non-shadow pixels with a power function. In the linear-correlation method the shadow was modeled as a combination of additive and multiplicative noise and then thenon-shadow pixels to the first order were restored by a linearfunction. In the histogram matching method the histogram ofthe shadowed region was matched to that of the non-shadow area of the same class in a window. In these algorithms theparameters were first calibrated before shadow removal byextracting training data sets from the image. The limitations inthese algorithms are that the estimated parameters can only beapplied in a local region and that the shadows in the trainingphase and in the estimation phase should be captured under thesame condition. The authors proposed a linear regression method to bridge non-shadow andshadow areas for eachclass in each band. Recently another linear regression basedmethod for shadow reconstruction has been proposed which assumed that both shadow and non-shadow pixels of eachclass follow a Gaussian distributionand then solved the linearregression parameters by the parametric estimation method.
The problem with these linear regression methods is that theylost local variability for each class due to the implementationin a global manner. In the algorithm proposed the first step was to collect ground truth region pairs for all classes i.e.non-shadow classes and their shadow counterparts and then these ground truth regions were utilized for supervised classification in shadow and non-shadow classes separately. In the shadow reconstruction method proposed in this project a similar ground truth collection procedure will be adopted but without the classification step.
An alternative shadow detection algorithm based on thresholding and morphological filtering together with an alternative shadow reconstruction algorithm based on the example learning method and Markov random field (MRF) is proposed in this project. During the shadow detection procedure the bimodal distributions of pixel values in the near infrared band and the panchromatic band are adopted for thresholding. During the shadow reconstruction procedure we model the relationship between non-shadow and the corresponding shadow pixels and between neighboring non-shadow pixels by employing MRF
1.1 MARKOV RANDOM FIELD
Markov random field in Markov network or undirectedgraphical model is a set of random variables having a Markov property described by an undirected graph. A Markov random field is similar to a Bayesian network in its representation of dependencies the differences being that Bayesian networks are directed and acyclic whereas Markov networks are undirected and may be cyclic. Thus a Markov network can represent certain dependencies that a Bayesian network cannot such as cyclic dependencies on the other hand it can't represent certain dependencies that a Bayesian network can such as induced dependencies.
When the probability distribution is strictly positive it is also referred to as a Gibbs random field because according to the Hamersley–Clifford theorem it can then be represented by a Gibbs measure. The prototypical Markov random field is the Ising model; indeed the Markov random field was introduced as the general setting for the Ising model. In the domain of artificial intelligence a Markov random field is used to model various low to mid level tasks in image processing and computer vision.For example MRF’s are used for image restoration image completion segmentation image registration texture synthesis super-resolution stereo matching and information retrieval.
Given an undirected graph G = (V E) a set of random variables X = (Xv)v ∈ V indexed by V form a Markov random field with respect to G if they satisfy the local Markov propertiesPairwise Markov property: Any two non-adjacent variables are conditionally independent given all other variables:
X_u ∐▒X_v | x_(v\{u ,v} )if{u ,v}∈E … (1.1)
Local Markov property: A variable is conditionally independent of all other variables given its neighbors:
X_v ∐▒X_(v\cl(v)) |X_(ne(v)) … (1.2)
Wherein (v) is the set of neighbors of v and cl (v) = {v} ∪ne (v) is the closed neighborhood of v.Global Markov property: Any two subsets of variables are conditionally independent given a separating subset:
X_(A∐▒X_B )| X_s …(1.3)
Where every path from a node in A to a node in B passes through S.The above three Markov properties are not equivalent to each other at all. In fact the Local Markov property is stronger than the Pairwise one while weaker than the Global one.
1.2 SHADOW DETECTION
For shadow detection we present an approach based on statistics of intensity we represent the picture in the YCbCr color space. We focus on the Y channel and compute itshistogram. Histogram dissension gives us a more contrast image at the Y channel.We compute the full average of the image at Y channelthen we perform sliding window iteration through the image. The sliding window size isreduced iteratively. In order to decide which pixels belong to the shadow we engage two approaches,
We consider being part of the shadow those pixels that have the intensity lower than 60% ofthe full average.
Compute the non-shadow point’s average for the sliding window. We consider being part ofthe shadow the pixels that have the intensity lower than the 70% of the window’s average .After this we compute the median filter to reduce the noise
As a further improvement we can introduce a segmentation algorithm and we can use our method foreach of the segments. The algorithm finds the lower intensities areas in the picture and this can be darker object or darkertextures not related to shadows or false detection. For such situations we can further decide if we have ashadow by taking into consideration that its border is smooth i.e. the border is usually rough for objects.
1.3 REMOTE SENSING
Remote sensing is the acquisition of information about an object or phenomenon without making physical contact with the object. In modern usage the term generally refers to the use of aerial sensor technologies to detect and classify objects on Earth both on the surface and in the atmosphere and oceans by means of propagated signals. There are two main types of remote sensing: passive remote sensing and active remote sensing. Passive sensors detect natural radiation that is emitted or reflected by the object or surrounding areas. Reflected sunlight is the most common source of radiation measured by passive sensors. Examples of passiveremotesensors include film photography infrared charge coupled devices and radiometers. Active collection on the other hand emits energy in order to scan objects and areas whereupon a sensor then detects and measures the radiation that is reflected or backscattered from the target. RADAR and LIDAR are examples of active remote sensing where the time delay between emission and return is measured establishing the location speed and direction of an object.
Remote sensing makes it possible to collect data on dangerous or inaccessible areas. Remote sensing applications include monitoring deforestation in areas such as the Amazon Basin glacial features in Arctic and Antarctic regions and depth sounding of coastal and ocean depths. Military collection during the Cold War made use of stand-off collection of data about dangerous border areas. Remote sensing also replaces costly and slow data collection on the ground ensuring in the process that areas or objects are not disturbed.
Orbital platforms collect and transmit data from different parts of the electromagnetic spectrum which in conjunction with larger scale aerial or ground based sensing and analysis provides researchers with enough information to monitor trends such as El Niño and other natural long and short term phenomena. Other uses include different areas of the earth sciences such as natural resource management agricultural fields such as land usage and conservation and national security and overhead ground based and standoff collection on border areas.
By satellite aircraft spacecraft buoy ship and helicopter images data is created to analyze and compare things like vegetation rates erosion pollution forestry weather and land use. These things can be mapped imaged tracked and observed. The process of remote sensing is also helpful for city planning archaeological investigations military observation and geomorphologic surveying.
1.4 APPLICATIONS OF REMOTE SENSING DATA
Conventional radar is mostly associated with aerial traffic control early warning and certain large scale meteorological data. Doppler radar is used by local law enforcements monitoring of speed limits and in enhanced meteorological collection such as wind speed and direction within weather systems. Other types of active collection include plasmas in the ionosphere. Interferometer synthetic aperture radar is used to produce precise digital elevation models of large scale terrain.
Laser and radar altimeters on satellites have provided a wide range of data. By measuring the bulges of water caused by gravity they map features on the seafloor to a resolution of a mile or so. By measuring the height and wavelength of ocean waves the altimeters measure wind speeds and direction and surface ocean currents and directions.
Light detection and ranging (LIDAR) is well known in examples of weapon ranging laser illuminated homing of projectiles. LIDAR is used to detect and measure the concentration of various chemicals in the atmosphere while airborne LIDAR can be used to measure heights of objects and features on the ground more accurately than with radar technology. Vegetation remote sensing is a principal application of LIDAR.
Radiometers and photometers are the most common instrument in use collecting reflected and emitted radiation in a wide range of frequencies. The most common are visible and infrared sensors followed by microwave gamma ray and rarely ultraviolet. They may also be used to detect the emission spectra of various chemicals providing data on chemical concentrations in the atmosphere.
Stereographic pairs of aerial photographs have often been used to make topographic maps by imagery and terrain analysts in trafficability and highway departments for potential routes.
Simultaneous multi-spectral platforms such as Lands at have been in use since the 70’s. These thematic mappers take images in multiple wavelengths of electro i.e. magnetic radiation multi-spectral and are usually found on Earth observation satellites for example the Lands at program or the IKONOS satellite. Maps of land cover and land use from thematic mapping can be used to prospect for minerals detect or monitor land usage deforestation and examine the health of indigenous plants and crops including entire farming regions or forests.
Hyper spectral imaging produces an image where each pixel has full spectral information with imaging narrow spectral bands over a contiguous spectral range. Hyper spectral imagers are used in various applications including mineralogy biology defense and environmental measurements.
Within the scope of the combat against desertificationremote sensing allows to follow-up and monitor risk areas in the long term to determine desertification factors to support decision makers in defining relevant measures of environmental management and to assess their impacts.
Sonar: passive sonar listening for the sound made by another object active sonar emitting pulses of sounds and listening for echoes used for detecting ranging and measurements of underwater objects and terrain.
Seismograms taken at different locations can locate and measure earthquakes after they occur by comparing the relative intensity and precise timings.
To coordinate a series of large scale observations most sensing systems depend on the following platform location what time it is and the rotation and orientation of the sensor. High end instruments now often use positional information from satellite navigation systems. The rotation and orientation is often provided within a degree or two with electronic compasses. Compasses can measure not just azimuth i.e. degrees to magnetic north but also altitude degrees above the horizon since the magnetic field curves into the Earth at different angles at different latitudes. More exact orientations require gyroscopic aided orientation periodically realigned by different methods including navigation from stars or known benchmarks.
1.5 DIGITAL IMAGE PROCESSING
Image Processing is the science of manipulating an image. With the advent of digital cameras and their easy interoperability with computers the process of digital image processing has acquired an entire new dimension and meaning. Image processing works with the digital images to enhance distort accentuate or highlight inherent details in the image. The goal of each operation is to achieve some details or we can generalize by saying extracting information from the system.
The area of image analysis (also called image understanding) is in between image processing and computer vision. There are no clear-cut boundaries in the continuum from image processing at one end to computer vision at the other. However one useful paradigm is to consider three types of computerized processes in this continuum: low- mid- and high-level processes.
Low-level: processes involve primitive operations such as image preprocessing to reduce noise contrast enhancement and image sharpening. A low-level process is characterized by the fact that both its inputs and outputs are images.
Mid-level: processing on images involves tasks such as segmentation partitioning an image into regions or objects descriptions of those objects to reduce them to a form suitable for computer processing and classifications of individual objects. A mid-level process is characterized by the fact that its inputs generally are images but its outputs are attributes extracted from those images.
High-level: processing involves “making sense” of an ensemble of recognized objects as in image analysis and at the far end of the continuum performing the cognitive functions normally associated with vision.
Based on the preceding comments it shows that a logical place of overlap between image processing and image analysis is the area of image recognition of individual regions or objects in an image. Thus we call digital image processing encompasses processes whose inputs and outputs are images and in addition encompasses processes that extracts attributes from images up to and including recognition of individual objects.
1.5.1 Analog Image Processing
Our eyes see the objects. They pass the information to the brain. Brain can understand the image. Such a human image processing mechanism is called analog image processing.
1.5.2 Digital Image Processing
With the help of scanning devices our computer systems acquire the image store the image in memory in digital form process the image in various ways .This mechanism is called digital image processing.
1.5.3 Digital Image Representation
The form monochrome image refers to a two dimensional light function f(x,y) here x and y denote spatial coordinates and the value of any point (x,y) is proportional to the brightness of the image of that point. A digital image can be considered a matrix of row and column indices identify a point in the image and the corresponding matrix elements value identifies the gray level at that point. The elements of such a digital array are called pixels or picture elements. When x y and the amplitude values of f are all finite discrete quantities. The image is a digital image. The field of digital image processing refers processing of digital images by means of a digital computer. A digital image is composed of a finite number of elements each of which has a particular location and value. These elements are referred to as picture elements image elements and pixels. Pixel is the term most widely used to denote the elements of the digital image. Vision is the most advanced of our senses so it is not surprising that images play the single most important role in human perception. However unlike humans who are limited to the visual band of the electromagnetic (EM) spectrum imaging machines cover almost the entire EM spectrum ranging from gamma to radio waves. They can operate on images generated by sources that humans are not accustomed to associating with images. These include ultra sound electron microscopy and computer generated images. Thus digital image processing encompasses a wide and varied field of application.
1.6 FUNDAMENTAL STEPS IN DIGITAL IMAGE PROCESSING
Digital image processing encompasses a broad range of hardware software and theoretical underpinnings. The following are the fundamental steps in image processing shows in the block diagram for fundamental steps in Digital Image Processing.
1.6.1 Image Acquisition
The image acquisition is to acquire a digital image. To do so require an image sensor and the capability to digitize the signal produced by the sensor. The sensor could be a monochrome camera that produces an entire image of the problem domain every 1/30 sec and the imaging sensor could be a linescan camera that produces a single image line at that time. In this case the object’s motion past the line scanner produces a two dimensional image. If the output of the camera or the other imaging sensor is not already in the digital form an analog to digital converter digitizes it.
1.6.2 Preprocessing
The key function of preprocessing is to improve the image in ways that increase the chances for success of the other processes.
1.6.3 Segmentation
Segmentation procedures partition an image into its constituent parts or objects. In general autonomous segmentation is one of the most difficult tasks in digital image processing. A rugged segmentation procedure brings the process a long way toward successful solution of imaging problems that require object to be identified individually.
1.6.4 Representation and description
Representation and description almost always follow the output of the segmentation stage which usually is raw pixel data constituting either the boundary of a region or all the points in the region itself. The first decision that must be made is whether the data should be represented as a boundary or as a complete region. Boundary representation is appropriate when the focus is on external shape characteristic such as corners and inflections. Regional representation is appropriate when the focus on internal properties such as texture or skeletal shape.
1.6.5 Recognition
Recognition is the process that assigns a label to an object based on its descriptors. We conclude coverage of digital image processing with the development of methods for recognition of individual objects.
1.6.6 Knowledge base
Knowledge about the problem domain is coded into an image processing system in the form of a Knowledge database. This knowledge may be as simple as detailing regions of an image where the information of interest is known to be located thus limiting the search that has to be conducted in seeking that information. The Knowledge base also can be quiet complex such as an interrelated list of all major possible defects in materials inspection problem or an image database containing high resolution satellite images of a region in connection with change detection applications.
1.7 ELEMENTS OF DIGITAL IMAGE PROCESSING SYSTEMS
The elements of a general purpose system capable of performing the image processing operations generally perform image acquisition storage processing communication and display.
1.7.1 Image Acquisition
Two elements are required to acquire digital images. The first is a physical device that is sensitive to a band in the electromagnetic energy spectrum such as the X-ray ultraviolet visible or infrared bands and that produces an electrical signal output proportional to the level of energy sensed. The second called a digitizer is a device for converting the electrical output of the physical sensing device into digital form.
1.7.2 Storage
An 8-bit image of size 1024*1024 pixels requires one million bytes of storage. Digital storage for image processing applications falls into 3 principle categories. They are short term storage used during processing,on line storage relatively used for fast recall and archival storage. Storage is measured in bytes Kbytes Mbytes Gbytes and Tbytes.
1.7.3 Processing
Processing of digital images involves procedures that are usually expressed in algorithmic form. Thus with the exception of image acquisition and display most images processing function can be implemented in software. The only reason for specialized image processing hardware is the need for speed in some application or to overcome some fundamental computer limitations. For example an important application of digital imaging is low-light microscopy.
1.7.4 Communication
Communication in digital image processing primarily involves local communication between image processing systems and remote communication from one point to another typically in connection with the transmission of image data. Hardware and software for local communication are readily available for most computers.
1.7.5 Display
Monochrome and color TV monitors are the principle display devices used in modern image processing systems. Monitors are driven by the output of a hardware image display module the backplane of the host computer or as a part of the hardware associated with an image processor. The signals and the output of the display module can be fed into an image recording device that produces the hard copy.
A COMPARATIVE STUDY ON SHADOW COMPENSATION OF COLOR AERIAL IMAGES IN INVARIANT COLOR MODELS
An improved algorithm has been adopted in comparative study on shadow compensation of color aerial images in invariant color models by Victor J. D. Tsai in 2006. Urban color aerial images shadows cast by cultural features may cause false color tone loss of feature information shape distortion of objects and failure of conjugate image matching within the shadow area. This paper presents an automatic property-based approach for the detection and compensation of shadow regions with shape information preserved in complex urban color aerial images for solving problems caused by cast shadows in digital image mapping. The technique is applied in several invariant color spaces that decouple luminance and chromaticity including HIS, HSV, HCV, YIQ and YCbCr models. Experimental results from de-shadowing color aerial images of a complex building and a highway segment in these color models are evaluated in terms of visual comparisons and shadow detection accuracy assessments. The results show the effectiveness of the proposed approach in revealing details under shadows and the suitability of these color models in de-shadowing urban color aerial images
2. CHARACTERISTICS OFSHADOW AND REMOVAL OF ITS EFFECTS FOR REMOTE SENSING IMAGERY
An improved algorithm has been adopted by Fumio Yamazaki in 2009 in the effects of shadow in remote sensing imagery are investigated. The measurement of radiance in sunlit and shadowed areas was carried out to investigate the spectral characteristics of sunlight. Based on this observation it is found that the radiance ratio shadow/sunlit increases as the sunlight get weaker and the ratio is dependent on the wavelength of sunlight. The darkness of shadow is also found to vary depending on the surrounding condition. Thus the condition to restore a shadow free image depends on the spectral bands and the location even in one image. A Quick Bird image is then introduced and the spectral characteristics of sunlit and shadowed areas are investigated. Based on these observations a method to detect shadowed areas and restore the shadow free radiance for the multi-spectral bands is proposed. The effectiveness of the shadow correction method is demonstrated for the Quick Bird image. Disadvantageof the system is that the performance is too low.
3. ROAD DETECTION FROM HIGH-RESOLUTION SATELLITE IMAGES USING ARTIFICIAL NEURAL NETWORKS
An improved algorithm has been adopted by M. Mokhtarzade in 2006 about road detection from satellite images can be considered as a classification process in which pixels are divided into road and background classes and can be used as a criterion in road extraction process to discriminate between road and non-road pixels. Apart from the spectral information textural parameters and contextual information are usually used by human being in object recognition from images. Contributing texture information in the neural network input parameters seems to be an improving idea for road detection from satellite images. Different texture parameters show different aspects of textural behavior in a defined neighborhood of a given pixel. Artificial neural networks are found to be superior to several previous techniques due in part to their ability to incorporate both spectral and contextual information. In this paper Neural Networks are applied on high resolution satellite images for road detection. At first road detection has been performed using only spectral information. Then different texture parameters including contrast energy entropy and homogeneity are computed for each pixel using gray level co-occurrence matrix (GLCM) from source image and a pre-classified road raster map is produced. To optimize neural networks' functionality and to evaluate the impact of contributing texture parameters in road detection extracted texture parameters are integrated with the spectral information. Major disadvantage is that it takes a lot of time for the entire process.
4. SHADOW DETECTION AND RADIOMETRIC RESTORATION IN SATELLITE HIGH RESOLUTION IMAGES
An improved algorithm has been adopted inshadow detection and radiometric restoration in satellite high resolution images by PooyaSarbandi in 2004. Here a new transformation which enables us to detect boundaries of cast shadows in high resolution satellite images is introduced. The transformation is based on color invariant indices. Different radiometric restoration techniques such as Gamma Correction Linear-Correlation Correction and Histogram Matching are introduced in order to restore the brightness of detected shadow area. Optical satellite images are contaminated with shadow the Disadvantageisnot effective.
PROPOSED METHODOLOGY
3.1INTRODUCTION
In this project we propose an alternative shadow detection algorithm based on thresholding and morphological filtering together with an alternative shadow reconstruction algorithm based on the example learning method and Markov random field (MRF). During the shadow detection procedure the bi-modal distributions of pixel values in the near-infrared (NIR) band and the panchromatic band are adopted for thresholding. During the shadow reconstruction procedure we model the relationship between non-shadow and the corresponding shadow pixels and between neighboring non-shadow pixels by employing MRF.In the shadow detection stage an initial shadow mask is generated by the thresholding method and then the noise and wrong shadow regions are removed by the morphological filtering method. The example-based learning phase the shadow and the corresponding nonshadow pixels are first manually sampled from the study scene and then these samples form a shadow library and a nonshadow library which are correlated by a Markov random field (MRF). During the inference phase the underlying land-cover pixels are reconstructed from the corresponding shadow pixels by adopting the Bayesian belief propagation algorithm to solve the MRF. The proposed shadow detection algorithm can generate accurate and continuous shadow masks and also that the estimated nonshadow regions from the proposed shadow reconstruction algorithm are highly compatible with their surrounding nonshadow regions. The effects of the reconstructed image on the application of classification by comparing the classification maps of images before and after shadow reconstruction.
3.2 PROBLEM FORMULATION AND MOTIVATION
Shadows occur when objects occlude direct light from a source of illumination which is usually the sun. According to the principle of formation shadows can be divided into cast shadow and self-shadow. Cast shadow is formulated by the projection of objects in the direction of the light source self-shadow refers to the part of the object that is not illuminated.
For a cast shadow the part of it where direct light is completely blocked by an object is termed the umbra while the part where direct light is partially blocked is termed the penumbra. Because of the existence of a penumbra there will not be a definite boundary between shadowed and non-shadowed areas. At present most of the VHSR satellite sensors are designed with orbit type of sun synchronous and equatorial crossing time earlier in a day; this is because the atmosphere is generally clearer in the morning than later in the day. Accordingly we propose a new method of detecting and reconstructing shadows in VHSR satellite images. Because self-shadows usually have higher brightness than cast shadows this project focuses only on the cast shadows as is the case in previous literature dealing with shadows of VHSR satellite images. With the variation of acquisition conditions and the height of erected objects the penumbra cannot sometimes be neglected particularly when the brightness of the surrounding shadowed areas is intense. The penumbra effect will be handled by shadow edge compensation in the proposed shadow detection algorithm.
For a given shadow region R s on a VHSR image we seek to estimate the corresponding non-shadow region Rn.This problem can be formulated as a maximum a posteriori problem i.e. finding Rn that maximizes posterior probability P(R_n│R_s ) which can be expressed as
(R_n ) ̂= 〖arg〗_Rn max〖P(R_n│R_s )〗 … (3.1)
According to the Bayesian theorem P(R_n│R_s )can be expressed as
P(R_n│R_s )=P(Rs ,R_n)/ P(Rs ) … (3.2)
Where P ( Rs,Rn) is the joint probability of R s and R n and P ( R s) is a prior probability of R s. Since P (R s) is a constant over non-shadow pixels (1) can be equivalently written as
(R_n ) ̂= 〖arg〗_Rn maxP (Rs ,R_n) … (3.3)
The research indicated that the darkness of shadow depends heavily on the surrounding conditions. Therefore it is necessary to build multiple relationships for each non-shadow class and its shadowed counterpart in one scene. In this project this problem is treated by manually collecting non-shadow and shadow pixel pairs in different locations of one scene for each class during the training procedure. Because of the existence of human interaction one requirement for the proposed shadow reconstruction algorithm is that the shadowed areas in the studied VHSR satellite image are not so dark thereby enabling human eyes to distinguish the classes under shadowed pixels. In the case that the shadowed areas are so dark one possible solution is to find a reference image with little or without shadows to assist in the training procedure. Another problem that was usually found in previous shadow reconstruction algorithms is that the reconstructed shadow area may not be consistent with its neighborhood non-shadow areas. In this project we handle this problem by utilizing the belief propagation method to keepthe reconstructed shadow area consistent with its neighborhood non-shadow areas.
3.3 EXISTING SYSTEM
Shadow detection in VHSR images two main approaches are reported namely the model-based and the property-based. The model-based approaches obtain good performance in detecting a particular type of objects like buildings and vehicles. The property-based approaches make use of certain shadow properties in images such as brightness spectral characteristics and geometry. In the thresholding-based methods the shadow and non-shadow pixels are determined according to a predefined threshold level which usually can be set according to the bimodal distribution of image histogram. In the color-transformation-based methods the red–green–blue (RGB) color image is first transformed to a 3-D space. The shadow detection method first transformed the RGB image into HSV space and then derived a normalized saturation–value difference index (NSVDI) to identify shadows via thresholding. In the region-growing-based methods the seed points are first selected and then each of the pixels is assigned to a segment according to their distance from those regions to which they could potentially be assigned. The classification methods can also be employed for shadow detection because of the commonly possessed properties in shadowed areas. The authors have proposed to utilize the support vector machine (SVM) classification method for shadow detection in which a binary classification procedure was implemented in a supervised manner to derive a shadow-versus-non-shadow mask.
To reconstruct the detected shadowed areas three algorithms were introduced. The gamma correction method considered the shadow as a multiplicative source that corrupts the brightness of the underlying pixels and then built the relationship between shadow and non-shadow pixels with a power function. The linear-correlation method the shadow was modeled as a combination of additive and multiplicative noise and then non-shadow pixels to the first order were restored by a linear function. The histogram-matching method the histogram of the shadowed region was matched to that of the non-shadow area of the same class in a window. The authors proposed linear regression method to bridge non-shadow and shadow areas for each class in each band.
Disadvantage
These approaches are not enough to deal with the great diversity of geometric structures which usually exist in VHSR satellite images of urban areas.
They lost local variability for each class due to the implementation in a global manner.
Lower classification performance.