WanlingLiuQianDong - Revision history

imported>Student221: /* Conclusion */

2019-12-14T09:41:24Z

Conclusion

← Older revision		Revision as of 09:41, 14 December 2019
Line 303:		Line 303:

	== Conclusion ==		== Conclusion ==
	In this project, we compared traditional methods with machine learning methods, and for machine learning approach, we tried several models. ISP tools often need complex parameter tuning process and would introduce more noise and artifacts, while CNN models are more straightforward to use and produce images of good quality. Compared with Tao's CNN model, our CNN model is computationally cheaper and yields better results. Generally speaking, our model is able to produce competitive and practical results in a reasonably fast time (though the training time for our model is long, the testing time is very short ~~(0.3s per image on average)~~ which means we can deblur images quickly after training on some dataset).		In this project, we compared traditional methods with machine learning methods, and for machine learning approach, we tried several models. ISP tools often need complex parameter tuning process and would introduce more noise and artifacts, while CNN models are more straightforward to use and produce images of good quality. Compared with Tao's CNN model, our CNN model is computationally cheaper and yields better results. Generally speaking, our model is able to produce competitive and practical results in a reasonably fast time (though the training time for our model is long, the testing time is very short which means we can deblur images quickly after training on some dataset).

	== Reference ==		== Reference ==

imported>Student221: /* Conclusion */

2019-12-14T09:40:57Z

Conclusion

← Older revision		Revision as of 09:40, 14 December 2019
Line 303:		Line 303:

	== Conclusion ==		== Conclusion ==
	In this project, we compared traditional methods with machine learning methods, and for machine learning approach, we tried several models. ISP tools often need complex parameter tuning process and would introduce more noise and artifacts, while CNN models are more straightforward to use and produce images of good quality. Compared with Tao's CNN model, our CNN model is computationally cheaper and yields better results. Generally speaking, our model is able to produce competitive and practical results in a reasonably fast time (though the training time for our model is long, the testing time is very short which means we can deblur images quickly after training on some dataset).		In this project, we compared traditional methods with machine learning methods, and for machine learning approach, we tried several models. ISP tools often need complex parameter tuning process and would introduce more noise and artifacts, while CNN models are more straightforward to use and produce images of good quality. Compared with Tao's CNN model, our CNN model is computationally cheaper and yields better results. Generally speaking, our model is able to produce competitive and practical results in a reasonably fast time (though the training time for our model is long, the testing time is very short (0.3s per image on average) which means we can deblur images quickly after training on some dataset).

	== Reference ==		== Reference ==

imported>Student221: /* Conclusion */

2019-12-14T08:58:49Z

Conclusion

← Older revision		Revision as of 08:58, 14 December 2019
Line 303:		Line 303:

	== Conclusion ==		== Conclusion ==
	In this project, we compared traditional methods with machine learning methods, and for machine learning approach, we tried several models. ISP tools often need complex parameter tuning process and would introduce more noise and artifacts, ~~but~~ CNN models are more straightforward to use and produce images of good quality. Compared with Tao's CNN model, our CNN model is computationally cheaper and yields better results. Generally speaking, our model is able to produce competitive and practical results in a reasonably fast time (though the training time for our model is long, the testing time is very short which means we can deblur images quickly after training on some dataset).		In this project, we compared traditional methods with machine learning methods, and for machine learning approach, we tried several models. ISP tools often need complex parameter tuning process and would introduce more noise and artifacts, while CNN models are more straightforward to use and produce images of good quality. Compared with Tao's CNN model, our CNN model is computationally cheaper and yields better results. Generally speaking, our model is able to produce competitive and practical results in a reasonably fast time (though the training time for our model is long, the testing time is very short which means we can deblur images quickly after training on some dataset).

	== Reference ==		== Reference ==

imported>Student221: /* SSIM */

2019-12-14T08:00:38Z

SSIM

← Older revision		Revision as of 08:00, 14 December 2019
Line 73:		Line 73:
	:<math>SSIM(x,y) =\frac{(2\mu_x \mu_y + C_1) + (2\sigma_{xy} + C_2)}{(\mu_x^2 + \mu_y^2 + C_1) + (\sigma_x^2 + \sigma_y^2 + C_2)} </math>		:<math>SSIM(x,y) =\frac{(2\mu_x \mu_y + C_1) + (2\sigma_{xy} + C_2)}{(\mu_x^2 + \mu_y^2 + C_1) + (\sigma_x^2 + \sigma_y^2 + C_2)} </math>

	~~These~~ three metrics above are commonly used for evaluating the quality of deblurring. Combining those three metrics we expect they can offer a general trend about the quality of results from the aspects of pixel-based difference, noise level, and structural difference. However, those parameters do not include much information related to human perception. Therefore, we investigated more about the metrics.		The above three metrics above are commonly used for evaluating the quality of deblurring. Combining those three metrics we expect they can offer a general trend about the quality of results from the aspects of pixel-based difference, noise level, and structural difference. However, those parameters do not include much information related to human perception. Therefore, we investigated more about the metrics.

	==== S-CIELAB ====		==== S-CIELAB ====

imported>Student221 at 08:00, 14 December 2019

2019-12-14T08:00:09Z

← Older revision		Revision as of 08:00, 14 December 2019
Line 59:		Line 59:

	=== Image Quality Evaluation===		=== Image Quality Evaluation===
	==== ~~Reference Metrics~~====		==== MSE ====
	The mean-square error computes the pixel-based difference between two images which represents the cumulative squared error between the deblurred and the original image. A lower value of MSE represents a lower error and a high image quality.		The mean-square error computes the pixel-based difference between two images which represents the cumulative squared error between the deblurred and the original image. A lower value of MSE represents a lower error and a high image quality.
	:<math>MSE = \frac {1}{mn}\sum_{i=0}^{m-1}\sum_{j=0}^{n-1}[I(i,j)-K(i,j)]^2</math>		:<math>MSE = \frac {1}{mn}\sum_{i=0}^{m-1}\sum_{j=0}^{n-1}[I(i,j)-K(i,j)]^2</math>

			==== PSNR ====
	PSNR is the ratio between the maximum possible value (power) of a signal and the power of distorting noise that affects the quality of its representation. The ratio can be used as a reference for image quality evaluation between the original and the deblurred image. An image with a higher PSNR has a higher image quality.		PSNR is the ratio between the maximum possible value (power) of a signal and the power of distorting noise that affects the quality of its representation. The ratio can be used as a reference for image quality evaluation between the original and the deblurred image. An image with a higher PSNR has a higher image quality.

	:<math>PSNR =10\log_{10}(\frac{MAX_I^2}{MSE})</math>		:<math>PSNR =10\log_{10}(\frac{MAX_I^2}{MSE})</math>

			==== SSIM ====
	SSIM is used for measuring the structural similarity between the two images. SSIM is designed to improve the evaluation metrics of images such as peak signal-to-noise ratio (PSNR) and mean squared error (MSE).		SSIM is used for measuring the structural similarity between the two images. SSIM is designed to improve the evaluation metrics of images such as peak signal-to-noise ratio (PSNR) and mean squared error (MSE).

Line 73:		Line 75:
	These three metrics above are commonly used for evaluating the quality of deblurring. Combining those three metrics we expect they can offer a general trend about the quality of results from the aspects of pixel-based difference, noise level, and structural difference. However, those parameters do not include much information related to human perception. Therefore, we investigated more about the metrics.		These three metrics above are commonly used for evaluating the quality of deblurring. Combining those three metrics we expect they can offer a general trend about the quality of results from the aspects of pixel-based difference, noise level, and structural difference. However, those parameters do not include much information related to human perception. Therefore, we investigated more about the metrics.

	==== ~~Perceptual Metrics~~ ====		==== S-CIELAB ====
	We also use S-CIELAB [5] to evaluate different models. S-CIELAB merges CIELAB and human spatial sensitivity and provides a difference between a reference and a corresponding test image.		We also use S-CIELAB [5] to evaluate different models. S-CIELAB merges CIELAB and human spatial sensitivity and provides a difference between a reference and a corresponding test image.
	The calculation step is shown in Fig. 8. For the S-CIELAB error measurement, the first step is to implement spatial filtering to the image data which simulates the human visual system, and then, the calculation is consistent with the basic CIELAB calculation.		The calculation step is shown in Fig. 8. For the S-CIELAB error measurement, the first step is to implement spatial filtering to the image data which simulates the human visual system, and then, the calculation is consistent with the basic CIELAB calculation.

imported>Student221: /* ISP */

2019-12-14T07:57:26Z

ISP

← Older revision		Revision as of 07:57, 14 December 2019
Line 11:		Line 11:
	== Methods ==		== Methods ==
	=== ISP ===		=== ISP ===
	The traditional way to process image and signal is to use ISP tools. ~~Processes~~ related with deblurring ~~are~~ commonly based on the evaluation of the point spread function and noise in the image. Algorithms are used to find the blur kernel and then recover the blurry image to the original image. In real circumstances, there would not be enough information for finding the blur kernel. Therefore, traditional deblurring methods are easy to suffer from intensive parameter tuning and poor performance.		The traditional way to process image and signal is to use ISP tools. Process related with deblurring is commonly based on the evaluation of the point spread function and noise in the image. Algorithms are used to find the blur kernel and then recover the blurry image to the original image. In real circumstances, there would not be enough information for finding the blur kernel. Therefore, traditional deblurring methods are easy to suffer from intensive parameter tuning and poor performance.

	[[File:ISP_pipeline.png\|thumb\|400px\|none\|Fig. 1 ISP pipeline (credit: Zheng Lyn)]]		[[File:ISP_pipeline.png\|thumb\|400px\|none\|Fig. 1 ISP pipeline (credit: Zheng Lyn)]]

imported>Student221: /* ISP */

2019-12-14T07:57:12Z

ISP

← Older revision		Revision as of 07:57, 14 December 2019
Line 11:		Line 11:
	== Methods ==		== Methods ==
	=== ISP ===		=== ISP ===
	The traditional way to process image and signal is to use ISP tools. Processes related with deblurring is commonly based on the evaluation of the point spread function and noise in the image. Algorithms are used to find the blur kernel and then recover the blurry image to the original image. In real circumstances, there would not be enough information for finding the blur kernel. Therefore, traditional deblurring methods are easy to suffer from intensive parameter tuning and poor performance.		The traditional way to process image and signal is to use ISP tools. Processes related with deblurring are commonly based on the evaluation of the point spread function and noise in the image. Algorithms are used to find the blur kernel and then recover the blurry image to the original image. In real circumstances, there would not be enough information for finding the blur kernel. Therefore, traditional deblurring methods are easy to suffer from intensive parameter tuning and poor performance.

	[[File:ISP_pipeline.png\|thumb\|400px\|none\|Fig. 1 ISP pipeline (credit: Zheng Lyn)]]		[[File:ISP_pipeline.png\|thumb\|400px\|none\|Fig. 1 ISP pipeline (credit: Zheng Lyn)]]

imported>Student221: /* GAN */

2019-12-14T07:53:11Z

GAN

← Older revision		Revision as of 07:53, 14 December 2019
Line 86:		Line 86:
	== Results ==		== Results ==
	=== GAN ===		=== GAN ===
	As for runtime consideration, we ~~use~~ 256x256 images for the preliminary experiments.		As for runtime consideration, we used 256x256 images for the preliminary experiments.
	* Model comparison		* Model comparison
	{\| class="Model Comparison" width="auto" style="margin-left: 0; margin-right: auto; border: none;"		{\| class="Model Comparison" width="auto" style="margin-left: 0; margin-right: auto; border: none;"

imported>Student221: /* CNN */

2019-12-14T07:51:48Z

CNN

← Older revision		Revision as of 07:51, 14 December 2019
Line 203:		Line 203:
	=== CNN ===		=== CNN ===

	We ~~compare~~ our preliminary CNN's result with GAN's result:		We compared our preliminary CNN's result with GAN's result:

	[[File:CNNvsGAN.png\|thumb\|none\|600px\|Fig. 9 CNN vs. GAN]]		[[File:CNNvsGAN.png\|thumb\|none\|600px\|Fig. 9 CNN vs. GAN]]

imported>Student221: /* GAN */

2019-12-14T07:51:24Z

GAN

← Older revision		Revision as of 07:51, 14 December 2019
Line 186:		Line 186:
	File:sharp.png\|Sharp Image		File:sharp.png\|Sharp Image
	File:Od_blurry.png\|Blurry Image(Object detection)		File:Od_blurry.png\|Blurry Image(Object detection)
	File:Od_unet.png\|GAN with ~~Unet~~(Object detection)		File:Od_unet.png\|GAN with U-net(Object detection)
	File:Od_vgg.png\|GAN with VGG16(Object detection)		File:Od_vgg.png\|GAN with VGG16(Object detection)
	File:Od_sharp.png\|Sharp Image(Object detection)		File:Od_sharp.png\|Sharp Image(Object detection)

← Older revision		Revision as of 07:51, 14 December 2019
Line 186:		Line 186:
	File:sharp.png\|Sharp Image		File:sharp.png\|Sharp Image
	File:Od_blurry.png\|Blurry Image(Object detection)		File:Od_blurry.png\|Blurry Image(Object detection)
	File:Od_unet.png\|GAN with ~~Unet~~(Object detection)		File:Od_unet.png\|GAN with U-net(Object detection)
	File:Od_vgg.png\|GAN with VGG16(Object detection)		File:Od_vgg.png\|GAN with VGG16(Object detection)
	File:Od_sharp.png\|Sharp Image(Object detection)		File:Od_sharp.png\|Sharp Image(Object detection)