Hybrid Video Coding. Thomas Wiegand: Digital Image Communication Hybrid Video Coding 1

Similar documents
CS 428: Fall Introduction to. Geometric Transformations (continued) Andrew Nealen, Rutgers, /20/2010 1

Lecture 10 Splines Introduction to Splines

Grade 11 Physics Homework 3. Be prepared to defend your choices in class!

Overview: motion-compensated coding

CS 314 Principles of Programming Languages

Jorge Salvador Marques, Stereo Reconstruction

CEE598 - Visual Sensing for Civil Infrastructure Eng. & Mgmt.

RULES OF DIFFERENTIATION LESSON PLAN. C2 Topic Overview CALCULUS

Procedure Abstraction

3D Image Representation through Hierarchical Tensor Decomposition, Based on SVD with Elementary Tensor of size 2 2 2

the marginal product. Using the rule for differentiating a power function,

Sound Insulation Analysis of Automotive Soundproof Material using Urethane and Rubber

Data Structures and Algorithms. The material for this lecture is drawn, in part, from The Practice of Programming (Kernighan & Pike) Chapter 2

4. Minimax and planning problems

Fisheye Lens Distortion Correction on Multicore and Hardware Accelerator Platforms

8.2 Areas in the Plane

EECS 487: Interactive Computer Graphics

Fill in the following table for the functions shown below.

CL-Path in B-Spline Form with Global Error Control for 3-Axis Sculptured Surface Machining

Giving credit where credit is due

DEFINITION OF THE LAPLACE TRANSFORM

Overview: motion estimation. Differential motion estimation

A High-Performance Area-Efficient Multifunction Interpolator

Systems & Biomedical Engineering Department. Transformation

called the vertex. The line through the focus perpendicular to the directrix is called the axis of the parabola.

WAYNESBORO AREA SCHOOL DISTRICT CURRICULUM - ACCELERATED GEOMETRY ACCELERATED. Should conjectures always be stated?

Algebra II Notes Unit Ten: Conic Sections

Parametric equations 8A

Overview Capturing Still Images Recording Videos Various Features Advanced Features

Image warping Li Zhang CS559

4.2 Implicit Differentiation

Maximum Flows: Polynomial Algorithms

Texture Mapping. Texture Mapping. Map textures to surfaces. Trompe L Oeil ( Deceive the Eye ) Texture map. The texture

Flow graph/networks MAX FLOW APPLICATIONS. Flow constraints. Max flow problem 4/26/12

The Laplace Transform

Camera Overview Capturing Still Images Recording Videos Various Features Advanced Features

USING ARTIFICIAL NEURAL NETWORKS TO APPROXIMATE A DISCRETE EVENT STOCHASTIC SIMULATION MODEL

geometric transformations

INFORMATION SECURITY

Energy/Performance Design of Memory Hierarchies for Processor-in-Memory Chips

Rou$ng. Rou$ng: Mapping Link to Path. Data and Control Planes. Rou$ng vs. Forwarding. Rou$ng Protocols. What Does the Protocol Compute?

M y. Image Warping. Targil 7 : Image Warping. Image Warping. 2D Geometric Transformations. image filtering: change range of image g(x) = T(f(x))

MTH 146 Conics Supplement

Nearest-Neighbor and Fault-Tolerant Quantum Circuit Implementation

Implementing Ray Casting in Tetrahedral Meshes with Programmable Graphics Hardware (Technical Report)

Interactive Rendering of Atmospheric Scattering Effects Using Graphics Hardware

Hybrid Signal-based and Geometry-based Prediction for Haptic Data Reduction

Camera Overview Capturing Still Images Recording Videos Various Features Advanced Features

Spline Curves. Color Interpolation. Normal Interpolation. Last Time? Today. glshademodel (GL_SMOOTH); Adjacency Data Structures. Mesh Simplification

Motion estimation. Announcements. Outline. Motion estimation

Dimmer time switch AlphaLux³ D / 27

See chapter 8 in the textbook. Dr Muhammad Al Salamah, Industrial Engineering, KFUPM

WORKSHOP 9 HEX MESH USING SWEEP VECTOR

Camera Overview Capturing Still Images Recording Videos Various Features Advanced Features

Curves & Surfaces. Last Time? Today. Readings for Today (pick one) Limitations of Polygonal Meshes. Today. Adjacency Data Structures

AlphaGo Overview. Overview. Digression: Zero sum, alternating move games. Original AlphaGo approach. Recent AlphaGo Zero approach 10/24/17

Video streaming over Vajda Tamás

Lighting and Shading

Image warping/morphing

CENG 477 Introduction to Computer Graphics. Modeling Transformations

Control Flow Analysis

Geometric Transformations Hearn & Baker Chapter 5. Some slides are taken from Robert Thomsons notes.

Describing motion. Unit 6.1. context. Fact File. Distance and displacement. Science

Describing Combinational circuits in BSV

Enhancement of Noisy Speech Using Sliding Discrete Cosine Transform

Chapter Six Chapter Six

Numerical Solution of ODE

Viewing Transformations I Comp 535

Programming (was due Friday) Last day to submit for late credit was yesterday. Today 6-7pm in AA204 All you need is a writing utensil and an ID

Principles of MRI EE225E / BIO265. Lecture 10. Instructor: Miki Lustig UC Berkeley, EECS. M. Lustig, EECS UC Berkeley

Projection & Interaction

Subband coding of image sequences using multiple vector quantizers. Emanuel Martins, Vitor Silva and Luís de Sá

CS559: Computer Graphics

REACTION ENUMERATION & CONDENSATION OF DOMAIN-LEVEL STRAND DISPLACEMENT SYSTEMS. Stefan Badelt DNA and Natural Algorithms (DNA) Group, Caltech

Frequency Domain Parameter Estimation of a Synchronous Generator Using Bi-objective Genetic Algorithms

Fig.1. Let a source of monochromatic light be incident on a slit of finite width a, as shown in Fig. 1.

Day-Night Long Range Camera System

Data Structures and Algorithms

STEREO PLANE MATCHING TECHNIQUE

MATH Differential Equations September 15, 2008 Project 1, Fall 2008 Due: September 24, 2008

In-HouseTrainingProgramme

Section 9.2 Hyperbolas

Wiley Plus. Assignment 1 is online:

Today. Quiz Introduction to pipelining

In fmri a Dual Echo Time EPI Pulse Sequence Can Induce Sources of Error in Dynamic Magnetic Field Maps

Diffraction I Version 1.0 : March 12, 2019

A METHOD OF MODELING DEFORMATION OF AN OBJECT EMPLOYING SURROUNDING VIDEO CAMERAS

6.8 Shortest Paths. Chapter 6. Dynamic Programming. Shortest Paths: Failed Attempts. Shortest Paths

Assignment 2. Due Monday Feb. 12, 10:00pm.

Mapping and Navigation

6.823 Computer System Architecture. Problem Set #3 Spring 2002

Order code M F Type of relay

Timers CT Range. CT-D Range. Electronic timers. CT-D Range. Phone: Fax: Web: -

EFFICIENT ON-LINE TESTING METHOD FOR A FLOATING-POINT ADDER

Vocabulary Check. 410 Chapter 4 Trigonometry

Try It. Implicit and Explicit Functions. Video. Exploration A. Differentiating with Respect to x

Projective geometry- 2D

Image interpolation. A reinterpretation of low-pass filtering. Image Interpolation

Exercises of PIV. incomplete draft, version 0.0. October 2009

Shortest Path Algorithms. Lecture I: Shortest Path Algorithms. Example. Graphs and Matrices. Setting: Dr Kieran T. Herley.

Transcription:

Hri Vieo Coing Principle of Hri Vieo Coing Moion-Compene Preicion Bi Allocion Moion Moel Moion Eimion Efficienc of Hri Vieo Coing Thom Wiegn: Digil Imge Communicion Hri Vieo Coing

I h een cuomr in he p o rnmi ucceive complee imge of he rnmie picure. [...] In ccornce wih hi invenion hi ifficul i voie rnmiing onl he ifference eween ucceive imge of he ojec. Thom Wiegn: Digil Imge Communicion Hri Vieo Coing

Hri Vieo Encoer [ ] u[ ] - Decoer Coer Conrol Inr-Frme DCT Coer Inr-Frme Decoer Conrol D DCT Coefficien u' [ ] Inr/Iner 0 ˆ [ ] Moion- Compene Preicor Moion Eimor ' [ ] Moion D Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 3

Hri Vieo Decoer Conrol D DCT Coefficien Decoer Inr-Frme Decoer u' [ ] Inr/Iner 0 Moion- Compene Preicor ' [ ] Moion D ˆ [ ] Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 4

Moion-Compene Preicion Moving ojec Sionr ckgroun Previou frme Δ Curren frme Diplce ojec ime Preicion for he luminnce ignl [] wihin he moving ojec: ˆ [ ] ( Δ from: Giro Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 5

Moion-Compene Preicion: Emple Frme [-] (previou Frme [] (curren Priion of frme ino lock (chemic Frme wih iplcemen vecor Difference eween moioncompene preicion n curren frme u[] Reference lock in frme Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 6

Re-conrine Moion Eimion Bi-re R Moion vecor re R m D Preicion error re R u Opimum re-off: D R D m R u Diplcemen error vrince Diplcemen error vrince cn e influence vi Block-ize qunizion of moion prmeer Choice of moion moel from: Giro Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 7

Re-Conrine Coer Conrol Efficienc incree vi ing coing moe Wh pr of he imge houl e coe uing wh meho? Prolem cn e poe : minimize iorion D ujec o re conrin R c min { D }.. R<Rc Coer Conrol uing Lgrngin opimizion: Solve n unconrine minimizion prolem λ min { D R } [Evere III 963] [Shohm Gerho 988] [Chou Lookugh Gr 989] Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 8

Lgrngin Opimizion in Vieo Coing A numer of inercion re ofen neglece Temporl epenenc ue o DPCM loop Spil epenenc of coing eciion Coniionl enrop coing Re-Conrine Moion Eimion [Sullivn Bker 99]: min {D m λ R m } Diorion fer moion compenion Lgrnge prmeer Numer of i for moion vecor Re-Conrine Moe Deciion [Wiegn e l. 996]: min {D λ R} Diorion fer reconrucion Lgrnge prmeer Numer of i for coing moe Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 9

Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 0 Cmer Moel Cmer Moel Nurl cmer-view cene re projece from he 3-D worl ino he -D imge plne A cmer moel cn e ue o ecrie projecion len n mpling z Imge plne 3D Poiion Imge poiion Focl poin F z 3-D poiion Imge poiion Imge plne Perpecive projecion moel Orhogrphic projecion moel Cn e ue when F«Z for ll coniere poin Z Y X P Z Y X P F p F p Y Z F X Z F Y X

Moion Moel Moion in 3-D pce correpon o iplcemen in he imge plne Moion compenion in he imge plne i conuce o provie preicion ignl for efficien vieo compreion Efficien moion-compene preicion ofen ue ie informion o rnmi he iplcemen Diplcemen mu e efficienl repreene for vieo compreion Moion moel rele 3-D moion o iplcemen uming reonle rericion of he moion n ojec in he 3-D worl Moion Moel f ( f ( : locion in previou imge : locion in curren imge : vecor of moion coefficien : iplcemen Thom Wiegn: Digil Imge Communicion Hri Vieo Coing

Mhemicl moel: Rericion: Perpecive Moion Moel 3 3 c c c c Roion n cling of rigi o in 3-D pce u no rnlion Trnlion roion n cling of plnr pch in 3-D Avnge: correpon o perpecive projecion moel Divnge: hperolic moion funcion Thom Wiegn: Digil Imge Communicion Hri Vieo Coing

Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 3 Trnlionl moion moel 4-Prmeer moion moel: rnlion zoom (ioropic Scling roion in imge plne Affine moion moel: Prolic moion moel Cn lo e viewe Tlor epnion of perpecive moion moel Orhogrphic Moion Moel Orhogrphic Moion Moel 3 3 3 3 6 5 4 3 6 5 4 3

Impc of he Affine Prmeer 50 rnlion 00 50 0-50 -00-50 0 50 00 50-50 -00-50 50 50 00 50 0-50 -00-50 0 50 00 50-50 cling 00 50 0-50 -00-50 0 50 00 50-50 -00-00 -50-50 50 3 heering 00 50 0-50 -00-50 0 50 00 50-50 -00-50 Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 4

Impc of he Prolic Prmeer 50 4 00 50 0-50 -00-50 0 50 00 50-50 -00-50 50 00 50 0-50 -00-50 0 50 00 50-50 -00 5 50 00 50 0-50 -00-50 0 50 00 50-50 -00-50 -50 50 6 00 50 0-50 -00-50 0 50 00 50-50 -00-50 Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 5

Impc of he Perpecive Prmeer c 50 00 50 0-50 -00-50 0 50 00 50-50 -00-50 50 00 50 0-50 -00-50 0 50 00 50-50 -00-50 c 50 00 50 0-50 -00-50 0 50 00 50-50 -00-50 Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 6

Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 7 Differenil Moion Eimion Differenil Moion Eimion Aume mll iplcemen : Aperure prolem: everl oervion require Inccure for iplcemen > 0.5 pel muligri meho ierion Minimize Diplce frme ifference Horizonl n vericl grien of imge ignl S u Δ Δ Δ ( ( ( ( ˆ( ( ( ( min u B B

Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 8 Grien Grien-Be Affine Refinemen Be Affine Refinemen Diplcemen vecor fiel i repreene Cominion 3 3 iel em of liner equion: Sem cn e olve uing e.g. peuo-invere minimizing ( ( ( ( ( 3 3 u Δ B B u ( rg min 3 3 3 3 u

Principle of Blockmching Suivie ever imge ino qure lock Fin one iplcemen vecor for ech lock Wihin erch rnge fin e mch h minimize n error meure. erch rnge in previou frme S k- Block of curen frme S k Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 9

Blockmching: : Error Meure Men qure error (um of qure error B B SSE( [ ( ( Δ] Sum of olue ifference SAD( B B ( ( Δ Approimel me performnce SAD le comple for ome rchiecure Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 0

Blockmching: : Serch Sregie I Full erch All poile iplcemen wihin he erch rnge re compre. Compuionll epenive Highl regulr prllelizle Thom Wiegn: Digil Imge Communicion Hri Vieo Coing

Block mching: Serch Sregie III Compuionl Complei Block comprion D logrihmic full erch 8 69 69 Emple: m. horizonl vericl iplcemen 6 ineger-pel ccurc - for pecil vecor (6 - wor ce from: Giro Thom Wiegn: Digil Imge Communicion Hri Vieo Coing

Block Comprion Spee-Up Tringle inequliie for SAD n SSE S S ( S S S lock lock S k k S k k lock N k lock ( S k k S k lock numer of erm in um Sreg:. Compue pril um for lock in curren n previou frme. Compre lock e on pril um 3. Omi full lock comprion if pril um inice wore error meure hn previou e reul Performnce: > 0 pee up of full erch lock mching repore emploing. Sum over 66 lock. Row wie lock projecion 3. Column wie lock projecion (Lin Ti IEEE Tr. Commun. M 97 Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 3 k N lock S lock k S k lock S k

Su-pel Trnlionl Moion Moion vecor re ofen no rerice o onl poin ino he ineger-pel gri of he reference frme Tpicl u-pel ccurcie: hlf-pel n qurer-pel Su-pel poiion re ofen eime refinemen Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 4

Hior of Moion Compenion Inrfrme coing: onl pil correlion eploie Complei DCT [Ahme Nrjn Ro 974] JPEG [99] incree Coniionl replenihmen H.0 [984] (DPCM clr qunizion Frme ifference coing H.0 Verion [988] Moion compenion: ineger-pel ccure iplcemen H.6 [99] Hlf-pel ccure moion compenion MPEG- [993] MPEG-/H.6 [994] Vrile lock-ize (66 & 88 moion compenion H.63 [996] MPEG-4 [999] Vrile lock-ize (66 44 n muli-frme moion compenion H.6L [00] Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 5

PSNR [B] 35 40 38 36 34 3 30 8 Mileone in Vieo Coing Vrile lock ize (66 44 qurer-pel muli-frme moion compenion (H.6L 00 Vrile lock ize (66 88 (H.63 996 qurer-pel moion compenion (MPEG-4 998 Bi-re Reucion: 75% Ineger-pel moion compenion (H.6 99 Inrfrme DCT coing (JPEG 990 Hlf-pel moion compenion (MPEG- 993 MPEG- 994 Coniionl Replenihmen (H.0 0 00 00 300 Foremn 0 Hz QCIF 00 frme Re [ki/] Frme Difference coing (H.0 988 Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 6

SNR [B] 40 38 36 34 3 30 8 Mileone in Vieo Coing Vrile lock ize (66 44 qurer-pel muli-frme moion compenion (H.6L 00 Viul Comprion Vrile lock ize (66 88 (H.63 996 qurer-pel moion compenion (MPEG-4 998 Ineger-pel moion compenion (H.6 99 Inrfrme DCT coing (JPEG 990 Hlf-pel moion compenion (MPEG- 993 MPEG- 994 Coniionl Replenihmen (H.0 0 00 00 300 Frme Difference coing (H.0 988 Re [ki/] Foremn 0 Hz QCIF 00 frme Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 7

Foremn @ 00 ki/ 0 Hz Inrfrme DCT coing No moion compenion (JPEG 990 Vrile lock ize (66 44 qurer-pel muli-frme moion compenion (H.6L 00 Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 8

Summr Vieo coing hri of moion compenion n preicion reiul coing Lgrngin i-llocion rule pecif conn lope llocion o moion coefficien n preicion error Moion moel cn repreen vriou kin of moion In prcice: ffine or 8-prmeer moel for cmer moion rnlionl moel for mll lock Differenil meho clcule iplcemen from pil n emporl ifference in he imge ignl Block mching compue error meure for cnie iplcemen n fin e mch Spee up lock mching erch meho n clever pplicion of ringle inequli Hri vieo coing h een ricll improve enhnce moion compenion cpiliie Thom Wiegn: Digil Imge Communicion Hri Vieo Coing 9