PDF Archive

Easily share your PDF documents with your contacts, on the Web and Social Networks.

Send a file File manager PDF Toolbox Search Help Contact



47I18 IJAET0118742 v6 iss6 2724 2731 .pdf



Original filename: 47I18-IJAET0118742_v6_iss6_2724-2731.pdf
Author: Editor IJAET

This PDF 1.5 document has been generated by Microsoft® Word 2013, and has been sent on pdf-archive.com on 04/07/2014 at 08:03, from IP address 117.211.x.x. The current document download page has been viewed 435 times.
File size: 635 KB (8 pages).
Privacy: public file




Download original PDF file









Document preview


International Journal of Advances in Engineering & Technology, Jan. 2014.
©IJAET
ISSN: 22311963

A NOVEL PROCESSING CHAIN FOR SHADOW DETECTION
AND RECONSTRUCTION IN VHR IMAGES
R. K. Nale.1 and S. A. Shinde.2
1

2

M.E. Student in Comp. Deptt, VP COE, Baramati. Pune University.
Assistant Professor. Comp. Deptt, VP COE, Baramati. Pune University

ABSTRACT
Image may contain shadow, which can lead to serious problem for full exploitation of image. This paper
proposes A Novel Processing Chain to solve this problem. The main aim of this chain process is not only detect
shadow region form image but also remove shadow region and reconstruct shadow less image. In this chain
process, initially we classified shadow vs. nonshadow region by using binary classification with the help of
support vector machine (SVM). For proper edge detection of shadow and nonshadow region, we apply canny
edge detection technique followed by image imposing. Finally shadow reconstruction, calculate mean and
standard deviation for shadow and nonshadow region of image. Find out mean difference between shadow and
nonshadow part of image and apply this difference to shadow part of image by normalization with help of
standard deviation.

KEYWORD: Shadow Detection, Shadow Removal, Image enhancement, Image restoration.

I.

INTRODUCTION

1.1 Introduction
Very High-resolution (VHR) mean there are many more pixels per square inch than the lower. This
technique creates a new era in satellite image processing. Due to (VHR) image it is very easy to
distinguish each and every object of image. We can easily identify full detail of each object from
(VHR) image. Such as vehicles, road, different shape of buildings, land marks etc. as well as shadow.
For such kind image we required new change detection, analysis and classification techniques.
Shadow is creating, when any object lies in the way of light emerging source for e.g. sun. This
shadow brightness level is varied. They are having different color tone than non shadow part of that
image. To identifying position, height and other parameter of building, bridges etc., we can use this
shadow [1], [2]. Frequency of this kind of shadow application is very less. Most of the time shadow is
defined as undesired information of image. Shadow may cause, destroy the shape of object, merge
objects, and lose small object or present false color tone. Shadow present in image lead to erroneous
result for Classification, mapping, interpretation etc. image processing method. To avoid these
drawbacks, and to increase image exploitability, shadow detection and shadow reconstruction are
necessary. Tsunami in 2004, importance of getting shadow-free images is massive. Shadow-free
images in very short time help to take crucial and rapid action in rescue mission [3].
Shadow detection can used model-based approach. This approach needs information about the
scenario and the sensor before processing the method, but such knowledge is not available easily. Due
to this drawback, most of shadow detection algorithm used shadow-property based approach. This
approach used hue, saturation, brightness properties of shadow [4]. Optimization for hue constant
RGB sensor [5] and invariant color model, both the techniques are used for shadow detection. These

2724

Vol. 6, Issue 6, pp. 2724-2731

International Journal of Advances in Engineering & Technology, Jan. 2014.
©IJAET
ISSN: 22311963
techniques are on the bases of space color transformation and threshold estimator [6]. In novel
successive thresholding scheme [7], different color space is analyzed to detect shadow.
For reconstruct shadow areas, there three methods: linear correlation, gamma correction, histogram
matching [8]. When any image contains shadow, surface texture of that image does not change
radically. To reconstruct shadow area by using local gamma transformation, they apply contextual
texture analysis between a segment of shadow and its neighbors, on the basis kind of surface under
shadow. Another method also used after detection of shadow area, which is known as histogram
matching method [9]. In this method, we adjust HIS value in shadow region by local surrounding of
each shadow. In linear regression method [10], each spectral band is carried out to correct the shadow
effects. This method is base on spectral signature of the spectral bands.
The remainder of this paper is organized as follows. In Section II, the problem of the presence of
shadows in VHR images is formulated. Section III details our approach. Section IV shows
experimental results, Section V draws the conclusions, and section VI contains future work.

1.2 Related work
Shadow detection and removal from image related work as followed.
1. The Mask Pyramid-Based Shadow Removal method [1], they first identify shadowed and lit areas
on the same surface in the scene using an illumination-invariant distance measure. These areas are
used to estimate the parameters of an affine shadow formation model. A novel pyramid-based
restoration process is then applied to produce a shadow-free image, while avoiding loss of texture
contrast and introduction of noise. They account for varying shadow intensity inside the shadowed
region by processing it from the interior towards the boundaries. Finally, to ensure a seamless
transition between the original and the recovered regions they apply image in painting along a thin
border.
2. A Complete Processing Chain for Shadow Detection and Reconstruction method [3], the detection
and classification tasks are implemented by means of the state-of-the-art support vector machine
approach. A quality check mechanism is integrated in order to reduce subsequent misreconstruction
problems. The reconstruction is based on a linear regression method to compensate shadow regions by
adjusting the intensities of the shaded pixels according to the statistical characteristics of the
corresponding nonshadow regions. Moreover, borders are explicitly handled by making use of
adaptive morphological filters and linear interpolation for the prevention of possible border artifacts in
the reconstructed image.
3. Detecting and removal method [4], begins with a segmentation of the color image. It is then
decided if a segment is a shadow by examination of its neighboring segments. They use the method
introduced in Finlayson to remove the shadows by zeroing the shadow’s borders in an edge
representation of the image, and then re-integrating the edge using the method introduced by Weiss.
This is done for all of the color channels thus leaving a shadow-free color image. The present method
requires neither a calibrated camera nor multiple images.
4. They use Removing Shadows from Images method [6], set out in to derive a 1-d illumination
invariant Shadow-free image. They then use this invariant image together with the original image to
locate shadow edges. By setting these shadow edges to zero in an edge representation of the original
image, and by subsequently re-integrating this edge representation by a method paralleling lightness
recovery, They are able to arrive at our sought after full color, shadow free image.

II.

PROBLEM FORMULATION

In satellite VHR images, especially in metro city areas, shadows created by big tower, bridges, tree
etc may destroy the information of image. Missing information having direct effect on common
analysis and processing operation of image, this leads to inefficient classification. Shadows create
when objects lies in way of the direct light from the illumination source, usually the sun. Shadow is
divided into two major parts: cast and self shadow as shown in Fig. 1. The shadow cause by direct
Projection of light on object is called Cast shadow and shadow cause by diffuse light present in scene
is called self shadow. Self shadow is the part of object. For simplicity, our paper does not propose to
distinguish between cast and self shadow. We concentrate on cast shadow, which is present in most of

2725

Vol. 6, Issue 6, pp. 2724-2731

International Journal of Advances in Engineering & Technology, Jan. 2014.
©IJAET
ISSN: 22311963
images. Cast shadow having property of homogeneous dark areas, which lead to loss of information in
image. This paper works on recover that information.

Figure 1:- Shadow of image

In this paper, we classified shadow and non shadow part of image. We applied canny edge detector
algorithm to detect edge of each part of image. For classification of each part, we use image imposing
technique with the help of support vector machine properties. Due to these properties, we can find out
shadow part and non shadow part. For reconstruction of shadow, we calculate mean and standard
deviation of shadow and non shadow part of image. Calculate mean difference between shadow and
non shadow part. Apply this difference on shadow pixel and we get reconstructed shadow less image.

III.

PROPOSED METHOD

Fig 2 shows a flowchart with the principal step methodology. Let us consider an image I of dimension
l×w. This image characterized by the presence of shadow area and composed of N bands. This
flowchart divide into three steps 1.Preprocessing 2.Shadow detection and 3. Shadows remove. In first
step, prepare image for image processing by converting color image into gray scale. After this step
image contain only intensity information. Apply binary classification on grayscale image in order to
distinguish shadow and nonshadow region. Second step is shadow detection, in this step we take
original image as an input and apply canny edge detection algorithm. Final result of this algorithm is
image having edges. After this, we apply image imposing process. Inputs for image imposing process
are, Image having edges and binary converted original image. Output of image imposing process is
shadow region identify in image. Final step of flow chart is shadow removal. For this step input is
shadow region identify image. Calculate mean and standard deviation for shadow as well as
nonshadow region of image. Calculate mean difference between shadow and nonshadow part of
image. Apply this difference on shadow part of image by normalization process with the help of
standard deviation.

2726

Vol. 6, Issue 6, pp. 2724-2731

International Journal of Advances in Engineering & Technology, Jan. 2014.
©IJAET
ISSN: 22311963

Figure 2:- Flow chart of proposed method.

3.1 Preprocessing
Preprocessing is divided into two steps, namely grayscale conversion followed by binary
classification.
3.1.1 Grayscale Conversion
The grayscale conversion is implemented by grayscale conversion algorithm, which convert color
image into grayscale image. Algorithm applies on original input image characterized by shadow
region. In this algorithm, first calculate length (l) and width (w) of image. Get pixel value in integer
format at (x,y) position of image, where x is the distance from the origin in the horizontal axis, y is the
distance from the origin in the vertical axis. Convert this integer value into hexadecimal value. By
doing this, we get Red(R), Green (G) and Blue (B) of that pixel. Then calculate GRAY value for that
pixel by using equation (1). Apply this calculated GRAY value to each Red(R), Green (G) and Blue
(B) value of that pixel i.e. R=GRAY, G=GRAY, B=GRAY. Now reset this new Red(R), Green (G)
and Blue (B) to that pixel. Apply same step for pixels from 0 to width (w) and for pixel from 0 to
length (l). Finally we get grayscale image.
(𝑅 + 𝐺 + 𝐵)
𝐺𝑅𝐴𝑌 =
(1)
3
3.1.2 Binary Classification
The binary classification is implemented by using support vector machine (SVM) properties, which is
useful for data classification in the literature. For image, we extract image bands and features from
image by using wavelet transform, which is useful for image classification. In this for each spectral
band, one-level stationary wavelet transform is applied. Output of this wavelet transform is four
space-frequencies features. By using symlet wavelet, we can obtain maximize sparseness of the
transformation while enforcing texture areas. Finally by using SVM, with the help of image band and
features, we can classify image into shadow vs. nonshadow classification.

2727

Vol. 6, Issue 6, pp. 2724-2731

International Journal of Advances in Engineering & Technology, Jan. 2014.
©IJAET
ISSN: 22311963
3.2 Shadow Detection
Shadow vs. nonshadow classification is the output of first step. But this classification not clears at the
edges of shadow and non shadow region, due to presence of noise in image. This noise is created by
slat and paper effect. So we apply second step, shadow Detection. Shadow Detection is divided into
two steps, namely edge detection followed by image imposing.
3.2.1 Edge Detection
The edge detection of image is implemented by Gaussian blur canny edge detection algorithm, which
detects the edges of image. Algorithm applies on original input image characterized by shadow
region. In this algorithm, first calculate length (l) and width (w) of image. Get pixel at (x,y) position
of image, where x is the distance from the origin in the horizontal axis, y is the distance from the
origin in the vertical axis. Calculate standard deviation (𝜎) for pixel, then Calculate Gaussian function
for pixel (x, y) by using equation (2).
𝑥 2 +𝑦 2
1

2𝜎 2
𝐺(𝑥, 𝑦) =
𝑒
(2)
2𝜋𝜎 2
Where x is the distance from the origin in the horizontal axis, y is the distance from the origin in the
vertical axis, and σ is the standard deviation of the Gaussian distribution. When applied in two
dimensions, this formula produces a surface whose contours are concentric circles with a Gaussian
distribution from the center point. Values from this distribution are used to build a convolution matrix
which is applied to the original image. Each pixel's new value is set to a weighted average of that
pixel's neighborhood. The original pixel's value receives the heaviest weight (having the highest
Gaussian value) and neighboring pixels receive smaller weights as their distance to the original pixel
increases. This results in a blur that preserves boundaries and edges better than other, more uniform
blurring filters. Final output of this step is image contains clear edge detection of shadow and
nonshadow region.
3.2.2 Image Imposing
This is the final step for shadow detection with clear boundary. Image imposing is procedure of the
placement of an image on top of another image. Usually use to add to the overall image effect. That
means we required two images. In our case, first image is the image get form edge detection step,
which is clear edges of shadow and nonshadow part and second image is binary converted image.
Apply image imposing on these two images and output is shadow detected region with clear edges.

3.3 Shadow Removing
Shadow removing is the final step of our chain process. In this shadow is removed in two steps. In
first step, we calculate mean and standard deviation for shadow and nonshadow region respectively.
In second step, calculate mean difference between shadow and nonshadow region and apply these
differences to shadow part.
3.3.1 Calculate Mean and Standard Deviation.
In this section we calculate mean for shadow region first. For mean calculation, first calculate length
(l) and width (w) of image. Get pixel value in integer format at (x,y) position of image, where x is the
distance from the origin in the horizontal axis, y is the distance from the origin in the vertical axis.
Convert this integer value into hexadecimal value. By doing this, we get Red(R), Green (G) and Blue
(B) of that pixel. Initially set red color total (𝑅𝑚 ), green color total (𝐺𝑚 ) and Blue color total (𝐵𝑚 ) to
null value. Add red color hexadecimal value one by one to find out total red color value 𝑅𝑚 for that
pixel, as show in equation (3). Same procedure applies for green color (𝐺𝑚 ) and blue color (𝐵𝑚 ) by
using equation no (4) and (5) respectively.
𝑅𝑚 = 𝑅𝑚 + 𝑅
(3)
𝐺𝑚 = 𝐺𝑚 + 𝐺
(4)
𝐵𝑚 = 𝐵𝑚 + 𝐵
(5)
Now calculate mean for red (µ𝑅 ), green (µ𝐺 ), blue (µ𝐵 ) by equation no (6), (7), (8). Apply same step
for pixels from 0 to width (w) and for pixel from 0 to length (l) of shadow region of image
𝑅𝑚
µ𝑅 =
(6)
(𝐿 ∗ 𝑊)

2728

Vol. 6, Issue 6, pp. 2724-2731

International Journal of Advances in Engineering & Technology, Jan. 2014.
©IJAET
ISSN: 22311963
𝐺𝑚
(7)
(𝐿 ∗ 𝑊)
𝐵𝑚
µ𝐵 =
(8)
(𝐿 ∗ 𝑊)
Finally we get mean for red (µ𝑅 ), green (µ𝐺 ), blue (µ𝐵 ) color of shadow region image. Now same
procedure applies to calculate mean for nonshadow region of image. After this calculate standard
deviation for shadow and nonshadow region of image.
3.3.2 Calculate Mean Difference.
After calculating mean color value for shadow and nonshadow region, now calculate mean difference
(D) between shadow and non shadow part by using equation (9).
𝐷 = |µ1 − µ2|
(9)
Then apply this (D) on R, G and B of Non shadow part of image by using normalization with the help
of standard deviation. Finally we got shadow free image.
µ𝐺 =

IV.

EXPERIMENTAL RESULT

To evaluate the performance of chain process, following image was used. As show in figure 3, (a)
initially we accept shadow contain image as input. After that we convert color image to grayscale
image (b). At (c) we detect edges of input image. Then we find out intermediate image at (d) and
finally resulted image (e), which is shadow free image.

Figure 3:- Reconstruction result (a) Original image. (b) Grayscale image. (c) Edge detected image. (d)
Intermediate image. (e) Resulted image.

V.

CONCLUSION

This paper solved the important problem of shadow content image. We solve this problem by not only
detecting shadow but also remove shadows from image. In our paper we use binary classification for
shadow and nonshadow part of image. To get clear boundary, we first apply canny edge detection
algorithm and then apply image imposing process on binary image with canny edge detected image.
So the output of image imposing is we get classification of shadow and nonshadow part with clear
boundary. For reconstruction process, we calculate mean and standard deviation for shadow and
nonshadow part of image and apply mean difference between shadow and nonshdow part to shadow
part. To improve final result this difference applies by using normalization process with the help of
standard deviation.

VI.

FUTURE WORK

These future directions could be work. First, structure element of image could be having trouble by
means of automatic adaptation procedure. This procedure depends on the sensor resolution and the
penumbra width, which depends on the sun direction and the building heights. Second, the
reconstruction of shadow part depends on classification accuracy of image. The height derived from a
digital elevation model could be considered as an additional input feature to better discriminate
between the thematic classes. Finally, last future related to reconstruction problem with more
statistical models. Although they would increase the computational complexity, they would lead to a

2729

Vol. 6, Issue 6, pp. 2724-2731

International Journal of Advances in Engineering & Technology, Jan. 2014.
©IJAET
ISSN: 22311963
better fitting of the shadow and nonshadow classes, thus resulting in a potentially better reconstruction
quality.

ACKNOWLEDGEMENTS
I express great many thanks to Prof. Santosh A. Shinde, for his great effort of supervising and leading
me, to accomplish this fine work. To college and department staff, they were a great source of support
and encouragement. To my friends and family, for their warm, kind encourages and loves. To every
person gave us something too light my pathway, I thanks for believing in me.

REFERENCES
[1]

Yael Shor, Dani Lischinski, "The Shadow Meets the Mask Pyramid-Based Shadow
Removal," EUROGRAPHICS 2008 Volume 27 (2008), Number 2.
[2] Ruiqi Guo, Qieyun Dai, Derek Hoiem, "Paired Regions for Shadow Detection and Removal," IEEE
Transactions on Pattern Analysis and Machine Intelligence, vol. 35, no. 12, pp. 2956-2967, Dec. 2013,
doi:10.1109/TPAMI.2012.214
[3] Luca Lorenzi, Farid Melgani, Grégoire Mercier, "A Complete Processing Chain for Shadow
Detection and Reconstruction in VHR Images," ieee transactions on geosciences and remote
sensing, Manuscript received February 22, 2011.
[4]Zvi Figov, Yoram Tal, and Moshe Koppel “ Detecting and Removing Shadows,” International Journal
of Engineering Trends and Technology (IJETT) - Volume4Issue5- May 2013
[5] R. Achanta, A. Shaji, K. Smith, A. Lucchi, P. Fua, and S. Susstrunk. “Slic superpixels compared to
state-of-the-art superpixel methods”. IEEE TPAMI, 2012.
[6] Graham D. Finlayson, Steven D. Hordley, and Mark S. Drew, "Removing Shadows from Images,".
[7] Ashraful Huq Suny and Nasrin Hakim Mithila, “A Shadow Detection and Removal from a Single Image
Using LAB Color Space,” IJCSI International Journal of Computer Science Issues, Vol. 10, Issue 4, No
2, July 2013
[8] A. Sanin, C. Sanderson, B.C. Lovell., “Shadow Detection: A Survey and Comparative Evaluation of
Recent Methods.”Pattern Recognition, Vol. 45, No. 4, pp. 1684–1695, 2012.
[9] A. Massalabi, H. Dong-Chen, G. B. Benie, and E. Beaudry, “Detecting information under and from
shadow in panchromatic ikonos images of the city of Sherbrooke,” in Proc. IGARSS, Sep. 2004, vol. 3,
pp. 2000–2004.
[10] K. Kouchi and F. Yamazaki, “Characteristics of tsunami-affected areas in moderate-resolution satellite
images,” IEEE Trans. Geosci. Remote Sens., vol. 45, no. 6, pp. 1650–1657, Jun. 2007.
[11] G. Finlayson and S. Süsstrunk, “Optimization for hue constant RGB sensors,” in Proc. IS&T/SID 10th
Color Imag. Conf., 2002, vol. 10, pp. 343–348.
[12] N. Otsu, “A threshold selection method from gray level histograms,” IEEE Trans. Syst., Man, Cybern.,
vol. SMC-9, no. 1, pp. 62–66, Jan. 1979.
[13] K. L. Chung, Y. R. Lin, and Y. H. Huang, “Efficient shadow detection of color aerial images based on
successive thresholding scheme,” IEEE Trans. Geosci. Remote Sens., vol. 47, no. 2, pp. 671–682, Feb.
2009.
[14] P. Sarabandi, F. Yamazaki, M. Matsuoka, and A. Kiremidjian, “Shadow detection and radiometric
restoration in satellite high resolution images,” in Proc. IEEE IGARSS, Sep. 2004, vol. 6, pp. 3744–
3747.
[15] J. Su, X. Lin, and D. Liu, “An automatic shadow detection and compensation method for remote
sensed color images,” in Proc. 8th Int. Conf. Signal Process., Nov. 2006, vol. 2, pp. 1–4.
[16] F. Yamazaki, W. Liu, and M. Takasaki, “Characteristic of shadow and removal of its effects for remote
sensing imagery,” in Proc. IEEE IGARSS, Jul. 2009, vol. 4, pp. 426–429.

AUTHORS
Nale Rajesh Keshav received his B.E. degree in Information Technology engineering from
the Mumbai University, Mumbai, in 2006. He is currently working toward the M.E. degree
in Computer engineering from the University of Pune, Pune. He is working as an Assistant
Professor in Dept of Information Technology of SVPM’s COE, Malegaon, Pune University.
His research interests lies in Digital Image Processing.

2730

Vol. 6, Issue 6, pp. 2724-2731

International Journal of Advances in Engineering & Technology, Jan. 2014.
©IJAET
ISSN: 22311963
Shimd Santosh A. received his B.E. degree in computer engineering (First Class with
Distinction) in the year 2003 from Pune University and M. E. Degree (First Class with
Distinction) in Computer Engineering in 2010 from Pune University. He has 10 years of
teaching experience at undergraduate and postgraduate level. Currently he is working as
Assistant Professor in Department of Computer Engineering of VPCOE, Baramati, Pune
University. His autobiography has been published in Marquie’s Who’ Who, an International
Magazine of Prominent Personalities of the World in the year 2012. He is also a Life
Member of IACSIT and ISTE professional bodies. His research interests are Digital Image Processing & Web
Services.

2731

Vol. 6, Issue 6, pp. 2724-2731


Related documents


PDF Document 47i18 ijaet0118742 v6 iss6 2724 2731
PDF Document 14i18 ijaet0118683 v6 iss6 2439 2447
PDF Document 6n19 ijaet0319318 v7 iss1 59 65
PDF Document 11 553 937 1 sm
PDF Document ijetr2101
PDF Document 38i14 ijaet0514323 v6 iss2 903to912


Related keywords