luis herranz , jose´ m. martı´nez

18
On the use of hierarchical prediction structures for ef cient summary generation of H.264/AVC bitstreams Luis Herranz, Jose´ M. Martı´nez Image Communication 2009

Upload: burke-garner

Post on 03-Jan-2016

17 views

Category:

Documents


0 download

DESCRIPTION

On the use of hierarchical prediction structures for efficient summary generation of H.264/AVC bitstreams. Luis Herranz , Jose´ M. Martı´nez. Image Communication 2009. Outline. Introduction System Framework Skimming Unit (SU) Coding order Generation of the summary - PowerPoint PPT Presentation

TRANSCRIPT

Page 1: Luis  Herranz , Jose´ M.  Martı´nez

On the use of hierarchical prediction structures for efficient summary generation of H.264/AVC

bitstreams

Luis Herranz, Jose´ M. Martı´nez

Image Communication 2009

Page 2: Luis  Herranz , Jose´ M.  Martı´nez

Outline

• Introduction• System Framework– Skimming Unit (SU)– Coding order– Generation of the summary– Presentation of the summary

• Experiment

Page 3: Luis  Herranz , Jose´ M.  Martı´nez

Introduction

• MPEG-2 prediction structure

Page 4: Luis  Herranz , Jose´ M.  Martı´nez

Introduction

• Hierarchical prediction structure in H.264/AVC using P frames (delay 0)

Page 5: Luis  Herranz , Jose´ M.  Martı´nez

Introduction

• Hierarchical prediction structure in H.264/AVC using B frames (delay 8)

Page 6: Luis  Herranz , Jose´ M.  Martı´nez

System FrameworkAnalyzing input bitstream Generating the summarized bitstream

# frames selected

Skimming unit (SU)

-The selection of SUs-Temporal level

Page 7: Luis  Herranz , Jose´ M.  Martı´nez

Skimming Units

Skimming Unit (SU)

Temporal layerIndex

𝑆𝑠𝑜𝑢𝑟𝑐𝑒 𝐹𝑟𝑎𝑚𝑒

1

𝑆𝑠𝑜𝑢𝑟𝑐𝑒=(𝐺0𝑇 ,𝐺1

𝑇 ,…,𝐺𝑀−1𝑇 )

Page 8: Luis  Herranz , Jose´ M.  Martı´nez

Skimming Units

Overlapped SUs

Page 9: Luis  Herranz , Jose´ M.  Martı´nez

Generation of the Summary

• Bit stream adaptation guided by skimming parameters

Page 10: Luis  Herranz , Jose´ M.  Martı´nez

Presentation of the Summary

• Constant frame rate• Variable frame rate– Signaling the changes of frame rate– Using a predefined set of SPSs

Page 11: Luis  Herranz , Jose´ M.  Martı´nez

Summarization Modalities

• Image storyboard: • Video Skim: • Frame dropping• Fast preview

Frame selection Skimming curve

Page 12: Luis  Herranz , Jose´ M.  Martı´nez

Image Storyboard

• Clustered using K-means algorithm

Page 13: Luis  Herranz , Jose´ M.  Martı´nez

Video Skimming• Face detection• Moving median filter

(a) Frames with face detected

(b) Skimming parameters -- 32 frames per GOP

(c) Skimming parameters -- 2 frames per GOP

Page 14: Luis  Herranz , Jose´ M.  Martı´nez

Fast Preview

• Motion activity• Median filter

(a) Filtered activity

(b) Skimming parameters -- 32 frames GOP

(c) Skimming parameters -- 2 frames GOP

Page 15: Luis  Herranz , Jose´ M.  Martı´nez

Subjective Evaluation

Visual distortion Semantic distortion

Page 16: Luis  Herranz , Jose´ M.  Martı´nez

Performance Evaluation

• Coding efficiency and processing time

Page 17: Luis  Herranz , Jose´ M.  Martı´nez

Performance Evaluation

• Bit stream extraction– Low computational cost– Faster generation of the bit stream

Page 18: Luis  Herranz , Jose´ M.  Martı´nez

skim storyboard

Fast playback