180 likes | 369 Views
Post-processing of JPEG image using MLP. Fall 2003 ECE539 Final Project Report Data Fok. Overview. Introduction Approach Experiments & Results Conclusion Demo. Introduction. Increase demand on graphic usage Graphics: large file size JPEG compression blocking artifact
E N D
Post-processing of JPEG image using MLP Fall 2003 ECE539 Final Project Report Data Fok
Overview • Introduction • Approach • Experiments & Results • Conclusion • Demo
Introduction • Increase demand on graphic usage • Graphics: large file size • JPEG compression blocking artifact • Unpopularity of JPEG 2000 • Removal of JPEG artifact
Approach • Multi Layer Perception • 15 inputs (5 x 3) • 5 R,G,B gradients of the neighbor pixels close to the block border • 6 outputs (2 x 3) • 2 R,G,B different of the original image and the compressed image on the pixels next to the block border
Approach – cont. • First order polynomial fit • Use the 4 pixels closest to the block border to estimate the value on the 2 pixels next to the border • Use as a control experiment
Approach – cont. • Image quality evaluate by • Human eyes • Peak signal to noise ratio (PSNR)
Experiment & Result • Optimal MLP structure after testing • Structure: 15-5-6 • Learning rate = 0.01 • Momentum = 0.7
Experiment & Result – cont. • Expt #1: grayscale image • train and test with the same image JPEG (0.14 bpp) PSNR = 41.2044 (dB) MLP postprocessed PSNR = 40.2514 (dB)
Experiment & Result – cont. • Expt #2: color image • train and test with the same image JPEG (0.18 bpp) PSNR = 38.2464 (dB) MLP postprocessed PSNR = 37.9718 (dB)
Experiment & Result – cont. • Expt #3: grayscale image • train with a high bpp image, test with a low bpp image JPEG (0.085 bpp) PSNR = 39.5696 (dB) MLP postprocessed PSNR = 39.6552 (dB)
Experiment & Result – cont. • Expt #4: color image • train with a high bpp image, test with a low bpp image • Training JPEG image bit rate = 0.374 bpp JPEG (0.065 bpp) PSNR = 37.4064 (dB) MLP postprocessed PSNR = 37.3664 (dB)
Experiment & Result – cont. • Expt #5: • train with a high bpp grayscale image, test with a low bpp color image • Training JPEG image bit rate = 0.255 bpp MLP postprocessed PSNR = 37.4312 (dB) JPEG (0.065 bpp) PSNR = 37.4064 (dB)
Experiment & Result – cont. • Expt #6: • train with a high bpp color image, test with a low bpp grayscale image • Training JPEG image bit rate = 0.255 bpp JPEG (0.085 bpp) PSNR = 39.5696 (dB) MLP postprocessed PSNR = 39.125 (dB)
Conclusion • MLP can decrease blocking artifact from experiment #3 • High quality image training data is needed • Current MLP structure does not suit color image training data • Further Study on the MLP structure for color image
References • W. B. Pennebaker and J. L. Mitchell, (1992) JPEG Still Image Compression Standard. New York: Van Nostrand Reinhold. • Martin Boliek, Charilaos Christopoulos, Eric Majani, (2000) JPEG 2000 Image Coding System, ISO/IEC JTCI/SC29 WGI, http://www.jpeg.org/CDs15444.html • Guoping Qiu, (2000) MLP for Adaptive Postprocessing Block-Coded Images. IEEE Transactions On Circuits And Systems For Video Technology, Vol. 10, No. 8, December 2000