Post processing of jpeg image using MLP

Download as ppt, pdf

0 likes529 views

This document reports on a study using a multi-layer perceptron (MLP) neural network to reduce blocking artifacts in JPEG compressed images. The MLP takes gradients of pixel values near block borders as input and outputs differences between original and compressed pixel values to post-process images. Experiments showed the MLP can improve quality of grayscale images compressed to low bit rates when trained on high quality data, but did not significantly improve color images. Further research is needed on MLP structures for reducing artifacts in color images.

Engineering

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)






=
MSE
PSNR
255
log10 10
[ ]
( )2
,
2
),(ˆ),(
MN
yxIyxI
MSE
yx
∑ −
=

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
JPEG (0.065 bpp)
PSNR = 37.4064 (dB)
MLP postprocessed
PSNR = 37.4312 (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

This document discusses the JPEG image compression standard. It begins with an overview of what JPEG is, including that it is an international standard for compressing color and grayscale images up to 24 bits per pixel. The document then discusses the basic JPEG compression pipeline of encoding and decoding. It also outlines some of the major algorithms used in JPEG compression, including color space transformation, discrete cosine transform (DCT), quantization, zigzag scanning, and entropy coding. A key component discussed is the DCT, which converts image data into frequency domains and is useful for energy compaction in compression. The document concludes with noting implementations of JPEG and DCT in fields like image processing, scientific analysis, and audio processing.

JPEG Image CompressionHemanth Kumar Mantri

��

JPEG is a lossy image compression algorithm, not a file format. It uses a 4-step process to compress images: 1) transforming RGB to YCbCr color space, 2) applying a discrete cosine transformation to identify redundant data, 3) quantizing the remaining data, and 4) encoding the result to minimize storage requirements. Typical compression ratios are 10:1 to 20:1 without visible loss and up to 100:1 compression for low quality applications.

Next generation image compression standards: JPEG XR and AICTouradj Ebrahimi

��

This document discusses the JPEG XR image compression standard developed from Microsoft's HD Photo, highlighting its focus on high dynamic range (HDR) imaging and improvements over previous standards. It details the technical components such as the reversible lapped biorthogonal transform (LBT), quantization, coefficient prediction, and scanning techniques used in JPEG XR. Performance evaluations against other standards like JPEG and JPEG 2000 are also summarized, emphasizing compression efficiency and computational requirements.

Image CompressionForrester High School

��

Common image compression techniques include RLE, LZW, and CLUT. The most common image file formats, which often use these compression techniques, are GIF, TIFF, JPEG, and PNG. BMP files are generally uncompressed. RLE stores repeated pixels by only recording the color once along with the number of repeats, reducing file size. LZW compresses by storing repeated bit sequences in a table rather than writing them out each time. CLUT reduces colors to a limited palette, cutting the storage needed per pixel.

Jpeg image compression using discrete cosine transform a surveyIJCSES Journal

��

The document is a survey on JPEG image compression using the Discrete Cosine Transform (DCT) and discusses its significant role in digital image processing for efficient transmission and storage of images. It covers the principles, techniques, and criteria for assessing image compression, along with the process of JPEG compression which includes color space conversion, quantization, and entropy coding. Overall, it emphasizes the necessity and efficiency of compression methods to manage the vast amounts of digital imagery in today's technology.

Multimedia image compression standardsMazin Alwaaly

��

The document discusses image compression standards, focusing on the JPEG standard developed in 1992, which uses lossy compression techniques. It outlines the main steps in JPEG compression, including transformations, quantization, and entropy coding, emphasizing techniques like the Discrete Cosine Transform and run-length coding. Additionally, it briefly mentions the JPEG2000 standard, which aims to improve compression efficiency, support large images, and enhance transmission quality.

JPEGRajatKumar471

��

The document explains JPEG, a standardized lossy image compression method used to reduce the size of digital images while maintaining acceptable quality. It details the processes involved in compression, including color transformation, quantization, coding, and different modes of encoding. Additionally, it outlines the advantages, such as reduced file sizes and adjustable compression, as well as disadvantages like quality loss and lack of support for transparency.

Project presentation image compression by manish myst, ssgbcoetManish Myst

��

This document discusses different image compression formats including GIF, PNG, JPEG, and MNG. It provides details on each format such as the algorithms and applications used. GIF uses LZW lossless compression while JPEG uses lossy compression to reduce file sizes. PNG also uses lossless compression with DEFLATE and prediction algorithms. The objective of image compression is to reduce file sizes by eliminating redundant image data through either lossy or lossless compression methods.

Digital Image Processing: Image Enhancement in the Frequency DomainMostafa G. M. Mostafa

��

This document is a chapter from a textbook on digital image processing. It discusses the discrete Fourier transform (DFT) and its properties. It also covers various filtering techniques that can be performed in the frequency domain, including low-pass, high-pass, band-pass, and homomorphic filters using approaches like Gaussian, Butterworth, and ideal filters. Homework problems 4.9 and 4.12 are also mentioned at the end.

Digital Image Processing: Image Enhancement in the Spatial DomainMostafa G. M. Mostafa

��

This document discusses various image enhancement techniques in the spatial domain, including point operations, histogram equalization, and spatial filtering. Point operations include transformations like thresholding, negatives, power-law and gamma corrections that manipulate individual pixel intensities. Histogram equalization improves contrast by spreading out the most frequent intensity values. Spatial filtering techniques like smoothing, sharpening and edge detection use small filters to modify pixel values based on neighboring areas.

Seema dipseemakashyap15

��

This document provides an overview of digital image processing. It discusses what digital image processing is, provides a brief history, and outlines some of the key stages involved, including image acquisition, enhancement, restoration, morphological processing, segmentation, representation and description, object recognition, and compression. It also discusses some example applications like medical imaging, autonomous vehicles, traffic monitoring, and biometrics. The document uses images to illustrate different concepts and stages in digital image processing.

Comparison between JPEG(DCT) and JPEG 2000(DWT) compression standardsRishab2612

��

This document compares JPEG and JPEG 2000 image compression standards, highlighting the need for compression to reduce storage space and discussing various compression techniques like DCT and DWT. It details the advantages and disadvantages of each method, noting that DCT is utilized in JPEG for medium bit rates while DWT in JPEG 2000 offers better quality at lower bit rates but can introduce blurring. Ultimately, DWT avoids blocking artifacts seen in DCT but may have a lower quality at low compression rates.

Image CompressionParamjeet Singh Jamwal

��

The document discusses image compression techniques in electrical and electronics engineering, detailing lossy and lossless compression methods, specifically focusing on JPEG compression algorithms. It explains concepts like discrete cosine transform (DCT), quantization, and Huffman encoding, along with the practical applications and advantages of image processing. Additionally, it touches upon the comparison between discrete wavelet transform (DWT) and DCT, and lists various applications of image processing in fields like computer vision and medical imaging.

JpegAnju Anjujosepj

��

JPEG is a widely used image compression method that operates in both lossy and lossless modes, designed primarily for continuous-tone images. It works by transforming color images into a luminance/chrominance color space, applying discrete cosine transform, and quantizing frequency components to achieve compression. The standard allows for various parameters to adjust the loss of data during compression and supports multiple output formats.

Image enhancementDr INBAMALAR T M

��

Digital images can be enhanced in various ways to improve quality. There are three main categories of enhancement techniques: spatial domain, frequency domain, and combination methods. Spatial domain methods operate directly on pixel values using point processing or neighborhood filtering. Key spatial techniques include contrast stretching, thresholding, and histogram equalization. Frequency domain methods modify an image's Fourier transform. Common transformations include logarithmic, power-law, and piecewise linear functions, which can increase contrast or highlight certain grayscale ranges. Proper enhancement improves an image's features for desired applications.

Digital image processing short quesstion answersAteeq Zada

��

This document discusses several techniques for 2D spatial image filtering and background subtraction in digital image processing. It covers linear filtering, Gaussian filtering, frame differencing, running averages, and mixtures of Gaussian models. The key techniques are linear filtering using a kernel or mask, Gaussian filtering to smooth images, and running averages or mixtures of Gaussians to model the background pixels over time while adapting to changes in illumination, motion, or scene geometry.

//STEIM Workshop: A Vernacular of File FormatsRosa ɯǝukɯɐn

��

This document provides definitions and explanations of various file formats and compression techniques, including: - Lossless vs lossy data compression - Raster images and how color depth affects stored color values - Image glitching techniques like imagebending and databending - How bytes and binary files are structured - Components of image files like headers, channels, and interleaved vs non-interleaved formats - Explanations of formats like BMP, GIF, PNG, PSD, JPG, TIFF, and others

igarss2011.pptgrssieee

��

1. The document discusses using the Auto-Binomial model (ABM) and Tikhonov optimization for despeckling and information extraction from Synthetic Aperture Radar (SAR) images. 2. It presents the ABM model and evidence maximization approach for estimating texture parameters and obtaining a Maximum A Posteriori (MAP) estimate of the despeckled image. 3. Experimental results on real SAR images show that the proposed ABM with Tikhonov optimization achieves better despeckling compared to other methods in terms of mean-to-noise ratio and equivalent number of looks, while also preserving phase information and point features.

Image enhancement techniquesSaideep

��

The document discusses various image enhancement techniques in digital image processing. It describes point operations like image negative, contrast stretching, thresholding, brightness enhancement, log transformation, and power law transformation. Contrast stretching expands the range of intensity levels and can be done by multiplying pixels with a constant, using a transfer function, or histogram equalization. Thresholding converts an image to binary by assigning pixel values above a threshold to one level and below to another. Log and power law transformations compress high intensity values and expand low values to enhance an image. Matlab code examples are provided for each technique.

Digital Image Processing: Image RestorationMostafa G. M. Mostafa

��

This document discusses image restoration techniques for noise removal, including: - Spatial domain filtering techniques like mean, median, and order statistics filters to remove random noise. - Frequency domain filtering like band reject filters to remove periodic noise. - Adaptive filtering techniques where the filter size changes depending on image characteristics within the filter region to better handle impulse noise.

image compression pptShivangi Saxena

��

This document discusses image compression techniques. It begins by defining image compression as reducing the data required to represent a digital image. It then discusses why image compression is needed for storage, transmission and other applications. The document outlines different types of redundancies that can be exploited in compression, including spatial, temporal and psychovisual redundancies. It categorizes compression techniques as lossless or lossy and describes several algorithms for each type, including Huffman coding, LZW coding, DPCM, DCT and others. Key aspects like prediction, quantization, fidelity criteria and compression models are also summarized.

Digital image processing - Image Enhancement (MATERIAL)Mathankumar S

��

This document discusses various image enhancement techniques including contrast stretching, compression of dynamic range, histogram equalization, and histogram specification. It provides definitions and explanations of these concepts with examples. Histogram equalization aims to produce a linear histogram to enhance an image, while histogram specification allows specifying a desired output histogram. Local enhancement can be achieved by applying these histogram processing methods over small non-overlapping regions instead of globally to reduce edge effects.

Design of Image Compression Algorithm using MATLABIJEEE

��

This document summarizes an image compression algorithm designed using Matlab. It discusses various image compression techniques including lossy and lossless compression methods. Lossy compression methods like JPEG are suitable for natural images while lossless methods are preferred for medical or archival images. The document also describes the image compression process involving encoding, quantization, decoding and calculation of compression ratios and quality metrics like PSNR and SNR. Specific compression algorithms discussed include Block Truncation Coding and techniques that exploit coding, inter-pixel and psychovisual redundancies like DCT used in JPEG.

DEMOSAICING OF REAL LOW LIGHTING IMAGES USING CFA 3.0sipij

��

The document presents an investigation into the demosaicing performance of a new color filter array (CFA) pattern termed CFA 3.0, specifically designed for low light conditions. Through extensive experiments with real low light images, it was found that CFA 3.0 demonstrates lower noise levels than CFA 1.0 and comparable performance to CFA 2.0. The study aims to assess the applicability of CFA 3.0 for practical low light imaging by applying various demosaicing algorithms and evaluating their performance metrics.

Non-essentiality of Correlation between Image and Depth Map in Free Viewpoin...Norishige Fukushima

��

This document summarizes an experiment on the correlation between images and depth maps in free viewpoint image coding. The experiment found that when using an accurate depth map, there is no need to consider correlation between the image and depth map. Various image codecs and post-filtering techniques were tested, and the best results were achieved using a post-filter set without a joint filter. Future work could optimize bit allocation between coded images and depth maps.

Image Enhancement in Spatial DomainDEEPASHRI HK

��

The document discusses various image enhancement techniques, focusing on gray level transformations, histogram processing, and spatial filtering methods. It categorizes enhancement approaches into spatial and frequency domain methods, detailing operations like image negatives, log transformations, and histogram equalization. Additionally, the document outlines the processes of local enhancement and image averaging, emphasizing the importance of manipulating pixel values for better visual representation in specific applications.

Image compression standardskirupasuchi1996

��

This document discusses various image compression standards and techniques. It begins with an introduction to image compression, noting that it reduces file sizes for storage or transmission while attempting to maintain image quality. It then outlines several international compression standards for binary images, photos, and video, including JPEG, MPEG, and H.261. The document focuses on JPEG, describing how it uses discrete cosine transform and quantization for lossy compression. It also discusses hierarchical and progressive modes for JPEG. In closing, the document presents challenges and results for motion segmentation and iris image segmentation.

Image enhancementvsaranya169

��

This document discusses image enhancement techniques. It begins with an introduction that defines image restoration as an objective process to recover the original image using prior knowledge, while image enhancement is a subjective process that seeks to improve the visual appearance without restoring fidelity. Next, it describes common image enhancement operations like noise removal, contrast adjustment, and zooming. It then discusses noise models, types of noise including photoelectric, impulse, and structured noise. Finally, it introduces filtering techniques for noise reduction, including band reject filters, bandpass filters, and notch filters.

Rapid Prototyping for XR: Lecture 4 - High Level Prototyping.Mark Billinghurst

��

Industrial internet of things IOT Week-3.pptxKNaveenKumarECE

��

More Related Content

What's hot (20)

Digital Image Processing: Image Enhancement in the Frequency DomainMostafa G. M. Mostafa

��

Digital Image Processing: Image Enhancement in the Spatial DomainMostafa G. M. Mostafa

��

Seema dipseemakashyap15

��

Comparison between JPEG(DCT) and JPEG 2000(DWT) compression standardsRishab2612

��

Image CompressionParamjeet Singh Jamwal

��

JpegAnju Anjujosepj

��

Image enhancementDr INBAMALAR T M

��

Digital image processing short quesstion answersAteeq Zada

��

//STEIM Workshop: A Vernacular of File FormatsRosa ɯǝukɯɐn

��

igarss2011.pptgrssieee

��

Image enhancement techniquesSaideep

��

Digital Image Processing: Image RestorationMostafa G. M. Mostafa

��

image compression pptShivangi Saxena

��

Digital image processing - Image Enhancement (MATERIAL)Mathankumar S

��

Design of Image Compression Algorithm using MATLABIJEEE

��

DEMOSAICING OF REAL LOW LIGHTING IMAGES USING CFA 3.0sipij

��

Non-essentiality of Correlation between Image and Depth Map in Free Viewpoin...Norishige Fukushima

��

Image Enhancement in Spatial DomainDEEPASHRI HK

��

Image compression standardskirupasuchi1996

��

Image enhancementvsaranya169

��

Digital Image Processing: Image Enhancement in the Frequency DomainMostafa G. M. Mostafa

��

Digital Image Processing: Image Enhancement in the Spatial DomainMostafa G. M. Mostafa

��

Seema dipseemakashyap15

��

Comparison between JPEG(DCT) and JPEG 2000(DWT) compression standardsRishab2612

��

Image CompressionParamjeet Singh Jamwal

��

JpegAnju Anjujosepj

��

Image enhancementDr INBAMALAR T M

��

Digital image processing short quesstion answersAteeq Zada

��

//STEIM Workshop: A Vernacular of File FormatsRosa ɯǝukɯɐn

��

igarss2011.pptgrssieee

��

Image enhancement techniquesSaideep

��

Digital Image Processing: Image RestorationMostafa G. M. Mostafa

��

image compression pptShivangi Saxena

��

Digital image processing - Image Enhancement (MATERIAL)Mathankumar S

��

Design of Image Compression Algorithm using MATLABIJEEE

��

DEMOSAICING OF REAL LOW LIGHTING IMAGES USING CFA 3.0sipij

��

Non-essentiality of Correlation between Image and Depth Map in Free Viewpoin...Norishige Fukushima

��

Image Enhancement in Spatial DomainDEEPASHRI HK

��

Image compression standardskirupasuchi1996

��

Image enhancementvsaranya169

��

Recently uploaded (20)

Rapid Prototyping for XR: Lecture 4 - High Level Prototyping.Mark Billinghurst

��

Industrial internet of things IOT Week-3.pptxKNaveenKumarECE

��

20CE404-Soil Mechanics - �ݺ�ߣ Share PPTsaravananr808639

��

Mechanical Vibration_MIC 202_iit roorkee.pdfisahiliitr

��

Call For Papers - 17th International Conference on Wireless & Mobile Networks...hosseinihamid192023

��

17th International Conference on Wireless & Mobile Networks (WiMoNe 2025) will provide an excellent international forum for sharing knowledge and results in theory, methodology and applications of Wireless & Mobile computing Environment. Current information age is witnessing a dramatic use of digital and electronic devices in the workplace and beyond. Wireless, Mobile Networks & its applications had received a significant and sustained research interest in terms of designing and deploying large scale and high performance computational applications in real life. The aim of the conference is to provide a platform to the researchers and practitioners from both academia as well as industry to meet and share cutting-edge development in the field.

Tesla-Stock-Analysis-and-Forecast.pptx (1).pptxmoonsony54

��

System design handwritten notes guidanceShabista Imam

��

Complete University of Calculus :: 2nd editionShabista Imam

��

NEW Strengthened Senior High School Gen Math.pptxDaryllWhere

��

FUNDAMENTALS OF COMPUTER ORGANIZATION AND ARCHITECTUREShabista Imam

��

Deep Learning for Natural Language Processing_FDP on 16 June 2025 MITS.pptxresming1

��

This gives an introduction to how NLP has evolved from the time of World War II till this date through the advances in approaches, architectures and word representations. From rule based approaches, it advanced to statistical approaches. from traditional machine learning algorithms it advanced to deep neural network architectures. Deep neural architectures include recurrent neural networks, long short term memory, gated recurrent units, seq2seq models, encoder decoder models, transformer architecture, upto large language models and vision language models which are multimodal in nature.

Rapid Prototyping for XR: Lecture 6 - AI for Prototyping and Research Directi...Mark Billinghurst

��

Abraham Silberschatz-Operating System Concepts (9th,2012.12).pdfShabista Imam

��

Industry 4.o the fourth revolutionWeek-2.pptxKNaveenKumarECE

��

Stay Safe Women Security Android App Project Report.pdfKamal Acharya

��

Women’s security is a critical issue in today’s world and it’s very much needed for every individual to be acting over such an issue. This document describes a GPS based “Women Security System'' that provides the combination of GPS devices as well as provide alerts and messages with an emergency button trigger whenever somebody is in trouble They might not have so much time, all that they have to do is generate a distress emergency signal by shaking up their phone. Our system provides a realizable, cost effective solution to problem detection. Nowdays due to recently happened cases such as rape by drivers or colleagues, burglary etc., women security, especially women security has become the foremost priority of the world. System uses the Global Positioning System (GPS) technology to find out the location of women. The information of women's position provided by the device can be viewed on Google maps using Internet or specialized software. The companies are looking for-ward to the security problem and require a system that will efficiently evaluate the problem of women security working in night shifts, traveling alone. We focus on the proposed model that can be used to deal with the security issue of women using GPS based tracking systems.

Fatality due to Falls at Working at Heightssuserb8994f

��

Microwatt: Open Tiny Core, Big PossibilitiesIBM

��

Unit III_One Dimensional Consolidation theorysaravananr808639

��

Complete guidance book of Asp.Net Web APIShabista Imam

��

Unlock the full potential of modern web development with the Complete Guidance Book of ASP.NET Web API—your all-in-one resource for mastering RESTful services using Microsoft’s powerful ASP.NET Core framework. This book takes you on a step-by-step journey from beginner to expert, covering everything from routing and controllers to security, performance optimization, and real-world architecture.

Introduction to Natural Language Processing - Stages in NLP Pipeline, Challen...resming1

��

Rapid Prototyping for XR: Lecture 4 - High Level Prototyping.Mark Billinghurst

��

Industrial internet of things IOT Week-3.pptxKNaveenKumarECE

��

20CE404-Soil Mechanics - �ݺ�ߣ Share PPTsaravananr808639

��

Mechanical Vibration_MIC 202_iit roorkee.pdfisahiliitr

��

Call For Papers - 17th International Conference on Wireless & Mobile Networks...hosseinihamid192023

��

Tesla-Stock-Analysis-and-Forecast.pptx (1).pptxmoonsony54

��

System design handwritten notes guidanceShabista Imam

��

Complete University of Calculus :: 2nd editionShabista Imam

��

NEW Strengthened Senior High School Gen Math.pptxDaryllWhere

��

FUNDAMENTALS OF COMPUTER ORGANIZATION AND ARCHITECTUREShabista Imam

��

Deep Learning for Natural Language Processing_FDP on 16 June 2025 MITS.pptxresming1

��

Rapid Prototyping for XR: Lecture 6 - AI for Prototyping and Research Directi...Mark Billinghurst

��

Abraham Silberschatz-Operating System Concepts (9th,2012.12).pdfShabista Imam

��

Industry 4.o the fourth revolutionWeek-2.pptxKNaveenKumarECE

��

Stay Safe Women Security Android App Project Report.pdfKamal Acharya

��

Fatality due to Falls at Working at Heightssuserb8994f

��

Microwatt: Open Tiny Core, Big PossibilitiesIBM

��

Unit III_One Dimensional Consolidation theorysaravananr808639

��

Complete guidance book of Asp.Net Web APIShabista Imam

��

Introduction to Natural Language Processing - Stages in NLP Pipeline, Challen...resming1

��

Post processing of jpeg image using MLP

1. Post-processing of JPEG image using MLP Fall 2003 ECE539 Final Project Report Data Fok

2. Overview  Introduction  Approach  Experiments & Results  Conclusion  Demo

3. Introduction  Increase demand on graphic usage  Graphics: large file size  JPEG compression blocking artifact  Unpopularity of JPEG 2000  Removal of JPEG artifact

4. 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

5. Approach – cont.

6. 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

7. Approach – cont.  Image quality evaluate by  Human eyes  Peak signal to noise ratio (PSNR)       = MSE PSNR 255 log10 10 [ ] ( )2 , 2 ),(ˆ),( MN yxIyxI MSE yx ∑ − =

8. Experiment & Result  Optimal MLP structure after testing  Structure: 15-5-6  Learning rate = 0.01  Momentum = 0.7

9. 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)

10. 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)

11. 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)

12. 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)

13. 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 JPEG (0.065 bpp) PSNR = 37.4064 (dB) MLP postprocessed PSNR = 37.4312 (dB)

14. 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)

15. 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

16. Demo

17. 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

18. Q&A

�ݺ�ߣ

Post processing of jpeg image using MLP

Recommended

More Related Content

What's hot (20)

Recently uploaded (20)

Post processing of jpeg image using MLP