LamTangYu - Revision history

imported>Projects221: /* Appendix II - Work partition (if a group project) */

2013-03-20T05:44:25Z

Appendix II - Work partition (if a group project)

← Older revision		Revision as of 05:44, 20 March 2013
Line 403:		Line 403:
	= Appendix II - Work partition (if a group project) =		= Appendix II - Work partition (if a group project) =
	The work was split evenly among all group members.		The work was split evenly among all group members.
	<li>Matt: EM algorithm and similarity scores ~~</li>~~		<li>Matt: EM algorithm and similarity scores
	<li>Eric & Chongxuan: Thresholding techniques and testing scripts~~</li>~~		<li>Eric & Chongxuan: Thresholding techniques and testing scripts
	<li>Everyone contributed to the wiki ~~</li>~~		<li>Everyone contributed to the wiki

imported>Projects221: /* Appendix II - Work partition (if a group project) */

2013-03-20T05:43:58Z

Appendix II - Work partition (if a group project)

← Older revision		Revision as of 05:43, 20 March 2013
Line 402:		Line 402:

	= Appendix II - Work partition (if a group project) =		= Appendix II - Work partition (if a group project) =
	~~Brian~~ and ~~Bob gave the lectures. Jon mucked around on~~ the wiki.		The work was split evenly among all group members.
			<li>Matt: EM algorithm and similarity scores </li>
			<li>Eric & Chongxuan: Thresholding techniques and testing scripts</li>
			<li>Everyone contributed to the wiki </li>

imported>Projects221: /* Appendix I - Code and Data */

2013-03-20T05:39:22Z

Appendix I - Code and Data

← Older revision		Revision as of 05:39, 20 March 2013
Line 396:		Line 396:

	==Code==		==Code==
	[[File:~~CodeFile~~.zip]]		[[File:FinalCode.zip]]

	==Data==		==Data==

imported>Projects221: /* Future Work */

2013-03-20T05:05:28Z

Future Work

← Older revision		Revision as of 05:05, 20 March 2013
Line 386:		Line 386:
	:1. Evaluate the performance of the algorithm on images corrupted by known types of noise (eg: by blurring, adding white Gaussian noise, and or rescaling)		:1. Evaluate the performance of the algorithm on images corrupted by known types of noise (eg: by blurring, adding white Gaussian noise, and or rescaling)
	:2. Vary the neighborhood size (number of interpolation coefficients).		:2. Vary the neighborhood size (number of interpolation coefficients).
	:3.		:3. Using a more complex EM model where we do not assume independence between different color channels.

	= References - Resources and related work =		= References - Resources and related work =

imported>Projects221: /* Conclusions */

2013-03-20T05:02:29Z

Conclusions

← Older revision		Revision as of 05:02, 20 March 2013
Line 380:		Line 380:

	= Conclusions =		= Conclusions =
	We showed in this project that detecting the presence or absence of CFA interpolation is a reasonable indicator that a digital image has been tampered with. We built a general CFA detector that did not require images to have a particular CFA interpolation algorithm and performed reasonably well. However, performance dropped when images underwent any kind of JPEG compression. We also found that smaller window sizes (eg: 64 pixels vs. ~~128~~ pixels) led to better performance without increasing the algorithm run time (on the order of 1/2 minute).		We showed in this project that detecting the presence or absence of CFA interpolation is a reasonable indicator that a digital image has been tampered with. We built a general CFA detector that did not require images to have a particular CFA interpolation algorithm and performed reasonably well. However, performance dropped when images underwent any kind of JPEG compression. We also found that smaller window sizes (eg: 64x64 pixels vs. 128x128 pixels) led to better performance without increasing the algorithm run time (on the order of 1/2 minute).

	=== Future Work ===		=== Future Work ===
	~~One~~ could extend this project ~~to include~~ the performance of the algorithm on images corrupted by noise ~~like~~ by blurring, adding white Gaussian noise, and or rescaling.		Given more time, we could extend this project in a few ways.
			:1. Evaluate the performance of the algorithm on images corrupted by known types of noise (eg: by blurring, adding white Gaussian noise, and or rescaling)
			:2. Vary the neighborhood size (number of interpolation coefficients).
			:3.

	= References - Resources and related work =		= References - Resources and related work =

imported>Projects221: /* Conclusions */

2013-03-20T04:57:29Z

Conclusions

← Older revision		Revision as of 04:57, 20 March 2013
Line 380:		Line 380:

	= Conclusions =		= Conclusions =
	We showed in this project that detecting the presence or absence of CFA interpolation is a reasonable indicator that a digital image has been tampered with. We built a general CFA detector that did not require images to have a particular CFA interpolation algorithm and performed reasonably well. However, performance dropped when images underwent any kind of JPEG compression. We also found that smaller window sizes (eg: 64 pixels vs. 128 pixels) led to better performance.		We showed in this project that detecting the presence or absence of CFA interpolation is a reasonable indicator that a digital image has been tampered with. We built a general CFA detector that did not require images to have a particular CFA interpolation algorithm and performed reasonably well. However, performance dropped when images underwent any kind of JPEG compression. We also found that smaller window sizes (eg: 64 pixels vs. 128 pixels) led to better performance without increasing the algorithm run time (on the order of 1/2 minute).

	=== Future Work ===		=== Future Work ===

imported>Projects221: /* Test Set Results */

2013-03-20T04:55:16Z

Test Set Results

← Older revision		Revision as of 04:55, 20 March 2013
Line 377:		Line 377:
	Figure 1 is an example where our algorithm successfully detects the location of the tampering. The black parts of the image indicate the difference between the fake and real image whereas the red box indicates our algoirthm's estimate of where the tampering happended.		Figure 1 is an example where our algorithm successfully detects the location of the tampering. The black parts of the image indicate the difference between the fake and real image whereas the red box indicates our algoirthm's estimate of where the tampering happended.

	[[File:localization.png\|400px\|thumb\|center\|Figure 1: Localization of the tampering was ~~sucessful~~.]]		[[File:localization.png\|400px\|thumb\|center\|Figure 1: Localization of the tampering was successful.]]

	= Conclusions =		= Conclusions =

imported>Projects221: /* Conclusions */

2013-03-20T04:55:06Z

Conclusions

← Older revision		Revision as of 04:55, 20 March 2013
Line 380:		Line 380:

	= Conclusions =		= Conclusions =
	We showed in this project that detecting the presence or absence of CFA interpolation is a reasonable indicator that a digital image has been tampered with. We built a general CFA detector that did not require images to have a particular CFA interpolation algorithm and performed reasonably well. However, performance dropped when images underwent any kind of JPEG compression.		We showed in this project that detecting the presence or absence of CFA interpolation is a reasonable indicator that a digital image has been tampered with. We built a general CFA detector that did not require images to have a particular CFA interpolation algorithm and performed reasonably well. However, performance dropped when images underwent any kind of JPEG compression. We also found that smaller window sizes (eg: 64 pixels vs. 128 pixels) led to better performance.

	=== Future Work ===		=== Future Work ===

imported>Projects221: /* Test Set Results */

2013-03-20T04:54:59Z

Test Set Results

← Older revision		Revision as of 04:54, 20 March 2013
Line 377:		Line 377:
	Figure 1 is an example where our algorithm successfully detects the location of the tampering. The black parts of the image indicate the difference between the fake and real image whereas the red box indicates our algoirthm's estimate of where the tampering happended.		Figure 1 is an example where our algorithm successfully detects the location of the tampering. The black parts of the image indicate the difference between the fake and real image whereas the red box indicates our algoirthm's estimate of where the tampering happended.

	[[File:localization.png\|400px\|thumb\|center\|Figure 5: ~~2D FFT~~ of the ~~probability map of a CFA interpolated image block~~.]]		[[File:localization.png\|400px\|thumb\|center\|Figure 1: Localization of the tampering was sucessful.]]

	= Conclusions =		= Conclusions =

imported>Projects221: /* Test Set Results */

2013-03-20T04:54:33Z

Test Set Results

← Older revision		Revision as of 04:54, 20 March 2013
Line 373:		Line 373:
	<br>		<br>
	Total error: 25%		Total error: 25%


			Figure 1 is an example where our algorithm successfully detects the location of the tampering. The black parts of the image indicate the difference between the fake and real image whereas the red box indicates our algoirthm's estimate of where the tampering happended.

			[[File:localization.png\|400px\|thumb\|center\|Figure 5: 2D FFT of the probability map of a CFA interpolated image block.]]

	= Conclusions =		= Conclusions =