Matte-Based Restoration of Vintage Video 指導老師 : 張元翔 主講人員 : 鄭功運.

Slides:



Advertisements
Similar presentations
Active Shape Models Suppose we have a statistical shape model –Trained from sets of examples How do we use it to interpret new images? Use an “Active Shape.
Advertisements

CSCE643: Computer Vision Bayesian Tracking & Particle Filtering Jinxiang Chai Some slides from Stephen Roth.
Various Regularization Methods in Computer Vision Min-Gyu Park Computer Vision Lab. School of Information and Communications GIST.
Investigation Into Optical Flow Problem in the Presence of Spatially-varying Motion Blur Mohammad Hossein Daraei June 2014 University.
Spatial-Temporal Consistency in Video Disparity Estimation ICASSP 2011 Ramsin Khoshabeh, Stanley H. Chan, Truong Q. Nguyen.
Hongliang Li, Senior Member, IEEE, Linfeng Xu, Member, IEEE, and Guanghui Liu Face Hallucination via Similarity Constraints.
Motivation Application driven -- VoD, Information on Demand (WWW), education, telemedicine, videoconference, videophone Storage capacity Large capacity.
Jue Wang Michael F. Cohen IEEE CVPR Outline 1. Introduction 2. Failure Modes For Previous Approaches 3. Robust Matting 3.1 Optimized Color Sampling.
Image Matting and Its Applications Chen-Yu Tseng Advisor: Sheng-Jyh Wang
Bayesian Robust Principal Component Analysis Presenter: Raghu Ranganathan ECE / CMR Tennessee Technological University January 21, 2011 Reading Group (Xinghao.
Chapter 8-3 Markov Random Fields 1. Topics 1. Introduction 1. Undirected Graphical Models 2. Terminology 2. Conditional Independence 3. Factorization.
Computer Vision Optical Flow
Temporal Video Denoising Based on Multihypothesis Motion Compensation Liwei Guo; Au, O.C.; Mengyao Ma; Zhiqin Liang; Hong Kong Univ. of Sci. & Technol.,
Wavelet-Domain Video Denoising Based on Reliability Measures Vladimir Zlokolica, Aleksandra Piˇzurica and Wilfried Philips Circuits and Systems for Video.
3D Video Generation and Service Based on a TOF Depth Sensor in MPEG-4 Multimedia Framework IEEE Consumer Electronics Sung-Yeol Kim Ji-Ho Cho Andres Koschan.
Learning to Detect A Salient Object Reporter: 鄭綱 (3/2)
Boundary matting for view synthesis Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Computer Vision and Image Understanding 103 (2006) 22–32.
Motion Detection And Analysis Michael Knowles Tuesday 13 th January 2004.
MSU CSE 803 Stockman Linear Operations Using Masks Masks are patterns used to define the weights used in averaging the neighbors of a pixel to compute.
High-Quality Video View Interpolation
Abstract Extracting a matte by previous approaches require the input image to be pre-segmented into three regions (trimap). This pre-segmentation based.
Object Recognition Using Geometric Hashing
An Iterative Optimization Approach for Unified Image Segmentation and Matting Hello everyone, my name is Jue Wang, I’m glad to be here to present our paper.
MSU CSE 803 Linear Operations Using Masks Masks are patterns used to define the weights used in averaging the neighbors of a pixel to compute some result.
Jinhui Tang †, Shuicheng Yan †, Richang Hong †, Guo-Jun Qi ‡, Tat-Seng Chua † † National University of Singapore ‡ University of Illinois at Urbana-Champaign.
Quakefinder : A Scalable Data Mining System for detecting Earthquakes from Space A paper by Paul Stolorz and Christopher Dean Presented by, Naresh Baliga.
Image Morphing CSC320: Introduction to Visual Computing
HMM-BASED PSEUDO-CLEAN SPEECH SYNTHESIS FOR SPLICE ALGORITHM Jun Du, Yu Hu, Li-Rong Dai, Ren-Hua Wang Wen-Yi Chu Department of Computer Science & Information.
Tracking Pedestrians Using Local Spatio- Temporal Motion Patterns in Extremely Crowded Scenes Louis Kratz and Ko Nishino IEEE TRANSACTIONS ON PATTERN ANALYSIS.
MPEG MPEG-VideoThis deals with the compression of video signals to about 1.5 Mbits/s; MPEG-AudioThis deals with the compression of digital audio signals.
TP15 - Tracking Computer Vision, FCUP, 2013 Miguel Coimbra Slides by Prof. Kristen Grauman.
Video Mosaics AllisonW. Klein Tyler Grant Adam Finkelstein Michael F. Cohen.
Motion-Compensated Noise Reduction of B &W Motion Picture Films EE392J Final Project ZHU Xiaoqing March, 2002.
Outline What Neural Networks are and why they are desirable Historical background Applications Strengths neural networks and advantages Status N.N and.
Object Stereo- Joint Stereo Matching and Object Segmentation Computer Vision and Pattern Recognition (CVPR), 2011 IEEE Conference on Michael Bleyer Vienna.
High-Resolution Interactive Panoramas with MPEG-4 발표자 : 김영백 임베디드시스템연구실.
Multifactor GPs Suppose now we wish to model different mappings for different styles. We will add a latent style vector s along with x, and define the.
Fields of Experts: A Framework for Learning Image Priors (Mon) Young Ki Baik, Computer Vision Lab.
Feng Liu, JinjunWang,ShenghuoZhu (MM’08) University of Wisconsin-Madison, NEC Laboratories America, Inc. 第一組: 資訊四 B 黃彥達 資訊碩一 R 蔡旻光 網媒碩二.
CVPR2013 Poster Detecting and Naming Actors in Movies using Generative Appearance Models.
1 Markov random field: A brief introduction (2) Tzu-Cheng Jen Institute of Electronics, NCTU
Analyzing wireless sensor network data under suppression and failure in transmission Alan E. Gelfand Institute of Statistics and Decision Sciences Duke.
Segmentation of Vehicles in Traffic Video Tun-Yu Chiang Wilson Lau.
The Digital Revolution Changing information. What is Digital?  Discrete values used for  Input  Processing  Transmission  Storage  Display  Derived.
Visual Computing Computer Vision 2 INFO410 & INFO350 S2 2015
 Present by 陳群元.  Introduction  Previous work  Predicting motion patterns  Spatio-temporal transition distribution  Discerning pedestrians  Experimental.
Tracking with dynamics
Improving Image Matting using Comprehensive Sampling Sets CVPR2013 Oral.
Feng Liu, JinjunWang,ShenghuoZhu (MM’08) University of Wisconsin-Madison, NEC Laboratories America, Inc. 第一組: 資訊四 B 黃彥達 資訊碩一 R 蔡旻光 網媒碩二.
Speaker Min-Koo Kang March 26, 2013 Depth Enhancement Technique by Sensor Fusion: MRF-based approach.
Stereo Video 1. Temporally Consistent Disparity Maps from Uncalibrated Stereo Videos 2. Real-time Spatiotemporal Stereo Matching Using the Dual-Cross-Bilateral.
Performance Issues in Doppler Ultrasound 1. 2 Fundamental Tradeoffs In pulsed modes (PW and color), maximum velocity without aliasing is In pulsed modes,
Jianchao Yang, John Wright, Thomas Huang, Yi Ma CVPR 2008 Image Super-Resolution as Sparse Representation of Raw Image Patches.
CSCI 631 – Foundations of Computer Vision March 15, 2016 Ashwini Imran Image Stitching.
Introduction To Computational and Biological Vision Max Binshtok Ohad Greenshpan March 2006 Shot Detection in video.
New Features Added to Our DTI Package XU, Dongrong Ph.D. Columbia University New York State Psychiatric Institute Support: 1R03EB A1 June 18, 2009.
Local Stereo Matching Using Motion Cue and Modified Census in Video Disparity Estimation Zucheul Lee, Ramsin Khoshabeh, Jason Juang and Truong Q. Nguyen.
CSCI 631 – Foundations of Computer Vision March 15, 2016 Ashwini Imran Image Stitching Link: singhashwini.mesinghashwini.me.
Technological Uncanny K. S'hell, C Kurtz, N. Vincent et E. André et M. Beugnet 1.
Motivations Paper: Directional Weighted Median Filter Paper: Fast Median Filters Proposed Strategy Simulation Results Conclusion References.
Compressive Coded Aperture Video Reconstruction
Motion Detection And Analysis
Fast Preprocessing for Robust Face Sketch Synthesis
Iterative Optimization
Turning to the Masters: Motion Capturing Cartoons
Image and Video Processing
Linear Operations Using Masks
Lecture 7 Spatial filtering.
Lark Kwon Choi, Alan Conrad Bovik
Presentation transcript:

Matte-Based Restoration of Vintage Video 指導老師 : 張元翔 主講人員 : 鄭功運

outline Introduction..…………………………. 3 Prior Work……………………………..4 Our Approach To Video Restoration…..5 Result…………………………………..12

Introduction Film archives are worth preserving because of their historical and cultural values. Because of film deterioration over time, making it very fragile. Film deterioration and its attendant artifacts are in turn caused by aging and chemical decomposition, and improper storage and handling. Film artifacts can be categorized as - 1. film wear (scratches, film hair, blotches, dust, line jitter) 2. luminance-chrominance (color grading, color fading, and luminance flicker) 3. Image instability loss of resolution 4. noise contamination. Fig.1

Prior Work Some of the most common artifacts in vintage video are blotches and scratches. Most approaches assume the artifact areas have been totally corrupted; such areas are identified and replaced by pixels in neighboring frames. In addition, such approaches assume artifacts can be identified in the current frame using only the previous and next frames. Assuming the missing areas have been correctly identified,techniques have been proposed to fill them. Fig. 1. This source of information provides an important constraint on artifact removal that enables a restored image to closely match the actual scene. In our work, we make use of as much information that is available in the digitized video for restoration.

Our Approach To Video Restoration What complicates the digital restoration process is the generative models for different artifacts (e.g., lines versus blotches) are likely to be different. In our matting model, each pixel in the corrupted frames is assumed to be the linear interpolation of clean pixel and artifact colors. A : artifact colors α : the matte that indicates how much of the signal is in the observed data The darker the pixel (smaller α ), the more contaminated the observed pixel. Note that “*” denotes pixel-wise multiplication.

Given (typically 5 in our work) consecutive frames in video, our goal is to find an optimal linear combination of the true video and color artifact, together with a matte α in [0,1], so that (1) or, on a per pixel basis (2) with x being a 3-tuple indexing space and time. So, the true pixel color is given by (3) Before we describe our restoration inference framework, we first define, from (3) (4) as the hypothesized alpha-premultiplied true color. It can be computed given the hypotheses of and without getting unstable (due to division by zero). will become useful later in both the 3-D spatio-temporal CRF and its inference method

The dotted box (flow estimation) is optional if there is very little motion in the scene between frames. We apply affine global motion estimation at the earlier stages, and optionally apply local motion estimation for refinement.

Goal : recover the clean frames P, alpha maps α, artifact color maps A, and motion field D, given corrupted frames G. It can be generally formulated as – We extract A and α given D and G then treat our input video as a CRF, with each pixel as a node. Unfortunately, is generally intractable. So – we wish to maximize it. This has the advantage of recovering from fewer number of states.

U : the bias on alpha B : the smoothness potential. ω u 、 ω b : weights. (Note that the smoothness potentials depend on the observed data.) The bias term for α is - Smoothness potential -

E2 : encouraging the clean frames to be consistent spatially E3 : encouraging the clean frames to be consistent temporally to measure the spatial and temporal consistency among clean frames, we should be comparing the hypothesized true colors and by taking their difference: to avoid division by small (noisy) α ’s, we premultiply this term α y α x by to yield the difference term α x q y – α y q x. E4 : encourages α to be continuous, E5 : encourages the artifact color distribution A to be continuous : with ω 1 = 6.2 ω 2 = 1.5 ω 3 = 2.44 ω 4 = 0.01 ω 5 = 1.0

Given α and A, we can then attempt to recover P. However, it is unwise to compute directly Estimate the true color of a pixel using its immediate spatio-temporal neighborhood. The estimated restored color at x is –

Result

Advantage : 適用於任何的汙損回復,且將每個 frame 裡的所有資訊都用到以利於回復 Disadvantage : 非常耗時 – n : the number of pixels in 5 frames k : the total number of possible states (i.e., 22 or 33) for each pixel T : the number of iterations ~The End~