Pixel and Edge Based Lluminant Color Estimation

3245 words (13 pages) Essay

26th Mar 2018 Computer Science Reference this


Disclaimer: This work has been submitted by a university student. This is not an example of the work produced by our Essay Writing Service. You can view samples of our professional work here.

Any opinions, findings, conclusions or recommendations expressed in this material are those of the authors and do not necessarily reflect the views of UKEssays.com.

Pixel and Edge Based Lluminant Color Estimation for Image Forgery Detection

Shahana N youseph and Dr.Rajesh Cherian Roy



Digital images are one of the powerful tools for communication. So Image security is a key issue when use digital images. With the development of powerful photo-editing software, such as Adobe Photoshop Light room 4, Apple Aperture 3, Corel PaintShop Pro X5, GIMP 2.8, photo manipulation is becoming more common. In this paper mainly detecting forged peoples in images. The main idea for the detection is, different images are captured under different illuminant condition, when combining these image fragments from different images, it is difficult to match the illumination conditions. This inconsistency of illumination leads to forgery detection. The main contribution of this method of forgery detection is how illuminant color can be used as a clue for forgery detection. The proposed method will be able to detect forgery using Linear SVM classification, with 70%-75% of accuracy.


Pixel based illuminant color estimation; Edge based illuminant color estimation,


Every day, millions of digital documents are produced by a variety of devices. They are distributed by newspapers, magazines, websites and television etc. In all these information channels, images are a powerful way for communication. It is not difficult to use computer graphics and image processing techniques to manipulate or to forge images. Video footage, scanned images, as well as digital and analogue images can be the target for manipulations. From a forensics perspective, several changes in a photograph are widely acceptable for improve the quality of images, e. g. to enhance the contrast, denoise an image, or highlight important regions etc.Forensics Science is a department for criminal investigation in distinct areas such as digital forensics, analogue forensics, multimedia forensics, network forensics etc.Image Forgery is the process of creating doctored/fake images, with the development of advanced image processing software’s such as Adobe Photoshop Light room 4, Apple Aperture 3, Corel PaintShop Pro X5, GIMP 2.8 etc forgeries in images is easy process. Image Forgery detection is Active and Passive. Digital watermarking is an example of active. The passive image forgery detection is a blind approach, which means it does not have any prior knowledge of input image. There are various methods used for the checking of authenticity of images. In this paper the method used is based on illumination. When light fall on an object color of the object is reflected, depends on illuminant color/light color. Objects having different color in different illumination condition. So when we forge an image or making composite of various images it is very difficult to maintain the consistency of illumination. Illumination is one of the criteria for forgery detection. Some other criteria’s are used for passive image forgery detection such as, JPEG compression properties, Projective geometry, Chromatic aberration, Color filter array (CFA) and inter pixel correlation etc.

  1. Literature Survey


Detection of forgery



Sandeep Gholap and P. K. Bora(2008)[2]

Estimate the illuminant color of selected specular highlight regions.

weak specular highlights on

Objects are also sufficient.

Not perform well photographs involving the human skin as

the cluster of human skin colors

Christian Riess and Elli Angelopoulou


User select the region want investigate

Can be applied to real world images

Multiple illuminant present ,it is difficult

Xuemin Wu, Zhen Fang(2011)[13]

Manual selection of reference block.(illuminant color estimation)

Specular free.

Illuminant color based on image block. (not reliable).

Tiago José de Carvalho

Christian Riess-(2013)[15]

Illuminant color classification

Semi automatic approach

Multi illuminant images not apply

Table1. Illuminant Color Based methods

  1. Proposed Method


First step is cropping face of the input image. This proposed method is mainly detecting forged peoples in an image. The estimation of the illuminant color is error-prone and it is affected by the materials in the scene, the illuminant color estimates on objects of similar material exhibit a lower relative error.Thus,the illuminant color detection to skin, mainly to faces. Pigmentation is the most obvious difference in skin characteristics. Second step is illuminant colour estimation, explained in next section. Fourth step is generation of illuminant map. Image is segmented with graph cut segmentation. Illuminant color is estimated using static methods on each segmented output with same index number. Based on the estimated illuminant color, apply it for the segments with same index number. The resulting output will be RGB components. This coloured representation of image with R G B components is termed as Illuminant Map. Fifth step is shape and colour feature extraction. For shape feature HOG Edge feature is used. An edge of illuminant map is extracted using various edge detection methods. Histogram of Oriented Gradients of edge points. For colour feature extraction. Colour Moments feature is used. Moments with first and second moments are extracted. Last step is SVM classification. Classify the illumination for each pair of faces in an image as either consistent or inconsistent. Assuming all selected faces are illuminated by the same light source, Train the SVM with two class with one class is for forged image and other for original image.When testing operation performed based on the test feature value image is classify either forged or original.



Pixel Based Illuminant Color Estimation

Pixel values of the entire are taken for illuminant color estimation. In this methods focussed on low level features. Such as Grey World, Max-RGB, Shades of grey. Simple and less complex calculation is used for the estimation, with the help of some static variables. So it is also known as static illuminant color estimation.

  1. Grey World Hypothesis: In Grey World, Illuminant color is estimated from Average Pixel values of images. Under a neutral light source or white light source, Average reflectance of the entire image is achromatic (Having no colors), if any deviation from this condition is due to color of illumination. This average reflected color will be the color of the light source.
  2. Max-RGB Hypothesis: In Max-RGB, illuminant color estimated from maximum response of Red Green Blue (RGB) channel. Maximum response is obtained from perfect reflectance. A surface having perfect reflectance property will respond (reflect) for the full range of light colors it captures, when light incident on it. Then this reflected color is actually the color of light source.
  3. Shades of Grey: Grey world and the max-RGB illuminant color estimation in terms of Minkowski norm, is called shades of gray.

, If p=1 Grey World Estimation


If p=∞ Max-RGB Estimation

If p=6 Shades of Grey Estimation

Edge Based Illuminant Color Estimation

Edge based illuminant color estimation is use low or higher order derivatives. In this methods edges and colors towards illuminant direction. In order to accurately estimate color of light source is use the pixel and edge points that coincide the illuminant direction. Highlights produce such types of points. In edge based estimation contains Grey edge and Weighted Grey edge estimation are used. In Weighted grey edge methods, using some weighting fuction to the edges. For that classifying the edges based on the photometric properties, material edges (e.g. edges between objects and object-background edges), shadow/shading edges (e.g. edges caused by the shape or position of an object with respect to the light source) and specular edges (i.e. highlights).These edges perform better influence on illuminant estimation. In Weighted Grey edge methods computing weighted average of edge points. The iterative weighting scheme is proposed, and by assigning this weighting scheme in to the grey edge method, the color of the light source is estimated. Edge based illuminant color estimation mainly contain,

• First Order Grey Edge

• Second Order Grey Edge

• Weighted Grey Edge

  1. First Order Grey Edge: The pth Minkowski norm of the first derivative of the reflectance in a scene is estimated. Computed by,

  1. Second Order Grey Edge: The pth Minkowski norm of the second derivative of the reflectance in a scene is estimated.

  1. Weighted Grey-Edge: Weighted Grey-Edge algorithm is computed by assigning a weighting function to the illuminant estimate. This weighting function is estimated by classifying edges based on the photometric properties and an iterative edge weighting scheme is generated.

• Derivative order x: the assumption that the average of the illuminants is achromatic can be extended to the absolute value of the sum of the derivatives of the image.

• Minkowski norm p: instead of simply adding intensities or derivatives, respectively, greater robustness can be achieved by computing the p-th Minkowski norm of these values.

• Gaussian smoothing σ: to reduce image noise, one can smooth the image prior to processing with a Gaussian kernel of standard deviation.

Specular Edge Weighting scheme: Specular weighting scheme is the ratio of the energy in the specular variant versus the total amount of derivative energy. This ratio translates to the specular edge weighting scheme given by:

, where


  1. Results

To check the accuracy of forgery detection using SVM classifier with SVM is trained with 50 forged and 50 original images and SVM is tested using total of 50 images where 25 are original and 25 are composite images downloaded from different websites in the Internet.SVM is trained several times for several testing process. First set of forgery detection testing is done with various illuminant estimation methods such as Grey World, MAX-RGB, Shades Of Grey and Grey Edge First and Second Order and weighted grey edge with shape feature and color feature extraction separately. In shape feature called HOG Edge use various edge detection methods such as Canny, Roberts, Prewitt, and Sobel for the comparative study. And finally the combination of color moment and HOG Edge is tested for forgery detection. Confusion matrix is generated accuracy is calculated.



True Positive (TP) – input-Forged, Output-Forged

True Negative (TN)- input-not forged ,output-not forged

False Positive (FP) -input-forged, output-not forged

False Negative (FN)-input-not forged, output-forged


Sl no

Illuminant color estimation Methods

Accuracy- HOG Edge

Accuracy – Color Moment

Accuracy- HOG Edge & Color Moment


Grey World










Shades Of Grey





First Order Grey Edge





Second Order Grey Edge





Weighted Grey Edge




Table 2. Estimated Accuracy Of fogery detection with Various Illuminant Color Estimation Methods

From the above result, when using all static illuminant color estimation method for forgery detection Weighted grey edge peform well when compare with other methods. Feature extraction used is HOG Edge and Color moments features for shape and color feature extraction. If use one feature extraction method only get 50%-64% of accuracy. If use combined HOG Edge and Color Moments features accuracy is improved to 66%-74%.

  1. Conclusions

Presented a new method for detecting forged images of people using the illuminant color Estimation. Estimate the illuminant color using Pixel and Edge based Illuminant estimation method, and generation of illuminant map. Canny edge detector are used to obtain edges of illuminant map for the extraction of shape features using HOG Edge descriptor, which is used to get Histogram of oriented Gradients of edge points. For color feature extraction use color moments features. These two features are tested separately with different illuminant estimation method for the comparative study. Combination of these two features is also used for forgery detection for the comparative study.From the result it is clear that combined HOG Edge and color features get more accuracy than method used shape and color features separately.Accuracy is Estimated using SVM Classifier.The Combined feature extraction with Weighted grey edge testing process get 74% of accuracy. The proposed method requires only a minimum amount of human interaction and provides a crisp statement on the authenticity of the image. Additionally, it is a significant advancement in the exploitation of illuminant color as a forensic cue. Prior color-based work either assumes complex user interaction or imposes very limiting assumptions.


The accuracy of the classification can be improved by using adding more content based features. Use of training based illuminant color estimation also improves accuracy.


[1] Hany Farid ,â€-Image Forgery Detection [A survey]â€-, IEEE signal processing magazine March-


[2] C. Riess and E. Angelopoulou, ―Scene illumination as an indicator of image manipulation,â€- Inf.

Hiding, vol. 6387, pp. 66–80, 2010.

[3] Gajanan K. Birajdar ,Vijay H. Mankar ,â€-Digital image forgery detection using passiv techniques:

A surveyâ€-Elsevier 2013 .

[4] J. van de Weijer, T. Gevers, and A. Gijsenij, ―Edge-based color constancy,â€- IEEE Trans. Image

Process., vol. 16, no. 9, pp. 2207–2214, Sep. 2007.

[5] M. Johnson and H. Farid, ―Exposing digital forgeries by detecting inconsistencies in lighting,â€- in

Proc. ACM Workshop on Multimedia and Security, New York, NY, USA, 2005, pp. 1–10.

[6] Yingda Lv Xuanjing Shen Haipeng Chen, ―An improved image blind identification based on

inconsistency in light source directionâ€- in Springer Science+Business Media, LLC 2010

[7] M. Johnson and H. Farid, ―Exposing digital forgeries in complex lighting environments,â€- IEEE

Trans. Inf. Forensics Security, vol. 3, no. 2, pp. 450–461, Jun. 2007

[8] M. Johnson and H. Farid, ―Exposing digital forgeries through specular highlights on the eye,â€- in

Proc. Int. Workshop on Inform. Hiding, 2007, pp. 311–325.

[9] E. Kee and H. Farid, ―Exposing digital forgeries from 3-D lighting environments,â€- in Proc. IEEE

Int. Workshop on Inform. Forensics and Security (WIFS), Dec. 2010, pp. 1–6.

[10] W. Fan, K. Wang, F. Cayre, and Z. Xiong, ―3D lighting-based image forgery detection using

shape-from-shading,â€- in Proc. Eur. Signal Processing Conf. (EUSIPCO), Aug. 2012, pp. 1777–


[11] E. Kee and H. Farid, ―Exposing digital forgeries from 3-D lighting environments,â€- in Proc. IEEE

Int. Workshop on Inform. Forensics and Security (WIFS), Dec. 2010, pp. 1–6.

[12] S. Gholap and P. K. Bora, ―Illuminant colour based image forensics,â€- in Proc. IEEE Region 10

Conf., 2008, pp. 1–5.

[13] X.Wu and Z. Fang, ―Image splicing detection using illuminant color inconsistency,â€- in Proc.

IEEE Int. Conf. Multimedia Inform. Networking and Security, Nov. 2011, pp. 600–

[14] P. Saboia, T. Carvalho, and A. Rocha, ―Eye specular highlights telltales for digital forensics: A

machine learning approach,â€- in Proc. IEEE Int. Conf. Image Processing (ICIP), 2011, pp. 1937–


[15] C. Riess and E. Angelopoulou, ―Physics-based illuminant color estimation as an image semantics

clue,â€- in Proc. IEEE Int. Conf. Image Processing, Nov. 2009, pp. 689–692. [12] S. Gholap

and P. K. Bora, ―Illuminant colour based image forensics,â€- in Proc. IEEE Region 10 Conf.,

2008, pp. 1–5.

[16] K. Barnard, V. Cardei, and B. Funt, ―A comparison of computational color constancy

algorithms–Part I: Methodology and Experiments With Synthesized Data,â€- IEEE Trans. Image

Process., vol. 11, no. 9, pp. 972–983, Sep. 2002.

[17] K. Barnard, L. Martin, A. Coath, and B. Funt, ―A comparison of computational color constancy

algorithms – Part II: Experiments With Image Data,â€- IEEE Trans. Image Process., vol. 11, no. 9,

pp. 985–996, Sep. 2002.

[18] A. Gijsenij, T. Gevers, and J. van deWeijer, ―Computational color constancy: Survey and

experiments,â€- IEEE Trans. Image Process., vol. 20, no. 9, pp. 2475–2489, Sep

[19] P. F. Felzenszwalb and D. P. Huttenlocher, ―Efficient graph-based image segmentation,â€- Int. J.

Comput. Vis., vol. 59, no. 2, pp. 167–181, 2004

Cite This Work

To export a reference to this article please select a referencing stye below:

Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.
Reference Copied to Clipboard.

Related Services

View all

DMCA / Removal Request

If you are the original writer of this essay and no longer wish to have your work published on the UKDiss.com website then please:

Related Lectures

Study for free with our range of university lectures!