Introduction to my Research

46
Research Review Kuo-Yen Lo 羅羅羅 羅 羅羅羅羅 2013.4.18 Sato Laboratory, University of Tokyo

description

Personal Homepage http://www.hci.iis.u-tokyo.ac.jp/~kylo/

Transcript of Introduction to my Research

Page 1: Introduction to my Research

Research Review

Kuo-Yen Lo 羅国彦

ロ コウエン

2013.4.18Sato Laboratory, University of Tokyo

Page 2: Introduction to my Research

Short Curriculum Vitae

• Personal Homepagehttp://www.hci.iis.u-tokyo.ac.jp/~kylo/

• Period of past yearsUniversity (2006 – 2010)Internship (2010 – 2011)Research Assistant (2012 – 2013)

• Two main research topic: 1. 2D-to-3D conversion2. Photo Aesthetics.

• Wrap up

Page 3: Introduction to my Research

Short Curriculum Vitae

• Language: English (TOEIC 935), JLPT(N1), Chinese(Native)

• Programming: C/C++, Matlab, Java(android)Technique: SIFT/SURF/HOG, K-means, GMM, kNN, SVM, PCA/LDA/ITML, bad-of-visual word, bilateral filter.

• Have traveled to: USA, Korea.Want to travel to: China, Thailand, Spain

• Why Japan-- historical and cultural connection-- camera companies and electronics maker here-- founded by Panasonic Scholarship

Page 4: Introduction to my Research

Visual Cues Low-levelMathematics

Machine LearningPsychology

Computer Vision

Page 5: Introduction to my Research

Overview

2006NTU

2007UPenn

2008OpenCV

2009RoboticsContest

GenderRecognition

Contest

2012ResearchAssistant

ICPR2012

ACCV_w2012

>>2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Page 6: Introduction to my Research

National Taiwan University

2006NTU

2007UPenn

2008OpenCV

>>

People in Vision• Yi-Ping Hung(MM12, CVPR11, CHI11, UIST11)• Yung-Yu Chuang(CVPR12*3)• C.J. Lin (Libsvm)• H.T. Lin(ICML12, NIPS12, CVPR11

KDD12 Champion)• Winston H. Hsu(MM12*6)

33,0001928 B.C.

1.6World rank 80

Page 7: Introduction to my Research

Summer School in UPenn

2006NTU

2007UPenn

2008OpenCV

>>Summer Language ProgramUniversity of Pennsylvania

Page 8: Introduction to my Research

Join the Lab

2006NTU

2007UPenn

2008OpenCV

>>Biophotonics and Bioimaging Laboratory

Prof. Ta-Te LinOpenGLOpenCV

BorlandC++ Builder

Page 9: Introduction to my Research

Robotics Contest @ ASABE 2009

2009RoboticsContest

GenderRecognition

Contest

>>• Problem:

Detecting and Positioning the circular obstacle• Technique:

Graphical simulation(OpenGL), Sensor• Material:

Boe-Bot Toolkit, IR sensor, Ultrasonic sensor, Zigbee wireless communication

Page 10: Introduction to my Research

Robotics Contest @ ASABE 2009

2009RoboticsContest

GenderRecognition

Contest

>>

Please Visit the following link for viewing the video:

http://youtu.be/8EjON8Y2OJ0

Page 11: Introduction to my Research

Gender Recognition Contest

2009RoboticsContest

GenderRecognition

Contest

>>• Problem:

Recognize the gender with single face image• Technique:

Viola-Jones Face detector (OpenCV)Feature-based alignmentFalse-alarm check

Rotate the image 35 degreeto detect all possible tilt face

Page 12: Introduction to my Research

Gender Recognition Contest

2009RoboticsContest

GenderRecognition

Contest

>>• Problem:

Recognize the gender with single face image• Technique:

Viola-Jones Face detector (OpenCV)Feature-based alignmentFalse-alarm check

Use Eye detectorto wrap the face tountilt view.

Page 13: Introduction to my Research

Gender Recognition Contest

2009RoboticsContest

GenderRecognition

Contest

>>• Problem:

Recognize the gender with single face image• Technique:

Viola-Jones Face detector (OpenCV)Feature-based alignmentFalse-alarm check

Utilize Skin color model, Eye and Mouth detector to filter the false-positive result from the V-J face detector.

Page 14: Introduction to my Research

Gender Recognition Contest

2009RoboticsContest

GenderRecognition

Contest

>>• Problem:

Recognize the gender with single face image• Technique:

Viola-Jones Face detector (OpenCV)Feature-based alignmentFalse-alarm check

• Performance1.2 second per 480*320 image

• Result~85% face detection accuracy~75% gender recognition accuracyWin 3rd place among 20 teams (Taiwan and China). Bonus 60,0000yen.

Page 15: Introduction to my Research

Fish Recognition

2010ISMAB

NTUGraduate

Internship@ TV corp.

>>• Problem:

Tuna species recognition for fishery conservation and management

• Task:Detection and Classification

Bigeye

YellowfinAlbacore

3 spices are considered2011Navy

Page 16: Introduction to my Research

Fish Recognition>>

• Problem:Tuna species recognition for fishery conservation and management

• Task:Detection and Classification

Fish Image are capturedin certain lighting condition with measurement plate.Body part is smooth,makes it reflect light well.

72%

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Page 17: Introduction to my Research

Fish Recognition>>

• Problem:Tuna species recognition for fishery conservation and management

• Task:Detection and Classification

B Y AB 89 10 5Y 9 86 3A 10 9 81

84%

34% 72% 52% 58%

Confusion Matrix

(Head)(Abdomen)(Tail fin) (Tail)

Discriminate part!

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Page 18: Introduction to my Research

Yeh, Graduation!>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Page 19: Introduction to my Research

2D-to-3D conversion>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

• Problem:Generating 3D videofrom 2D content.

• Inspiration:3D information isrecovered by depthcues

Captured View + Depth

Page 20: Introduction to my Research

2D-to-3D conversion>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Reality Comfort

• Accurate depth map• Correct depth order• Real-time processing

• Clear boundary• Temporal smoothness• Visual impression

Page 21: Introduction to my Research

How people perceive depth?>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

1. Low-level cue 2. Scene Recognition

Page 22: Introduction to my Research

2D-to-3D conversion>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Video frame + motion estimation[ICCE 2009]

Approaches1. Depth map by motion2. Depth map by saliency 3. Depth map by prior

information fusion

Page 23: Introduction to my Research

2D-to-3D conversion>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy Video frame + Saliency map [SDA 2010]

Approaches1. Depth map by motion2. Depth map by saliency 3. Depth map by prior

information fusion

Page 24: Introduction to my Research

Introduction to Bilateral Filter>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Bilateral Filter [Tomasi, ICCV98]:

f(x) h(x)

“Bi” lateral = Spatial term + Range term

Page 25: Introduction to my Research

Introduction to Bilateral Filter>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Bilateral Filter [Tomasi, ICCV98]:

f(x) h(x)

“Bi” lateral = Spatial term + Range term

Page 26: Introduction to my Research

Introduction to Bilateral Filter>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Bilateral Filter [Tomasi, ICCV98]:

f(x) h(x)

“Bi” lateral = Spatial term + Range term

Page 27: Introduction to my Research

Application of Bilateral Filter>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

“Bi” lateral = Spatial term + Range termSmooth Target Edge-Preserving Result

And this one?

Page 28: Introduction to my Research

2D-to-3D conversion>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Prior fusion [Siggraph 2009]1. Decide Geometric perspective2. Integrate Image and Depth map by Bilateral filter

Page 29: Introduction to my Research

One-year in Navy>>

2010ISMAB

NTUGraduate

Internship@ TV corp.

2011Navy

Page 30: Introduction to my Research

Academia Sinica

2012ResearchAssistant

ICPR2012

ACCV2012

>>Institute of Information Science(Central Research Academy)

People in Vision• Chu-Song Chen (CVPR12, CVPR11*2)• Mark H. Liao (MM12*2, MM11*2)• Y.-C. Frank Wang (ECCV12, CVPR12)• Yen-Yu Lin (CVPR13, MM12, TPAMI11)

Prof. Chen

Page 31: Introduction to my Research

PHOTO AESTHETICS CLASSIFICATIONPredicting the visual appealing quality of photos

Page 32: Introduction to my Research

good?>>

jcar@DPChallenge

Page 33: Introduction to my Research

good?>>

Mnet @ DPChallenge

Page 34: Introduction to my Research

Which one is better?>>

Page 35: Introduction to my Research

Voted by online photo community>>

Average: 5.088 votes

Average: 7.292 votes

Page 36: Introduction to my Research

Reason?>>

Page 37: Introduction to my Research

Reason?>>

BoundaryAlternating repetition(Texture)

Contrast Levels of scale

Roughness

Strong centers

Positive space

Local symmetries

The Void

Not-separateness

Good shape

GradientsEchoes

Simplicity and Inner Calm

Deep interlock and ambiguity

Color

Composition

HarmoniumRichness

Page 38: Introduction to my Research

Application>>

Image Search & Management Photo evaluation system

Embedded Camera system Media analysis

Page 39: Introduction to my Research

Photo Aesthetics

2012ResearchAssistant

ICPR2012

ACCV2012

>>• Problem:

Recognition the appealing quality ofphoto by computational approaches.

• Technique:Image analysis, Pattern recognition,Crowdsourcing, Psychology, Photography

• Application

Page 40: Introduction to my Research

ICPR 2012

2012ResearchAssistant

ICPR2012

ACCV2012

>>As a Pattern Recognition Problem…Comparison of feature 1. Edge distribution, Color histogram,

Hue, Saturation.. [Ke, CVPR06]2. SIFT + BOV [Marchesotti , ICCV11]3. Composition layout (Edge + HSV),

Color palette, contrast.. [Proposed]Result

Item Speed on PC Accuracy

CVPR06 0.2s 81%

ICCV11 4s 85%

Proposed 0.16s 84%

[ Photo aesthetics assessment with efficiency ]

Page 41: Introduction to my Research

Extraction of Color Information

Extract N Dominant colors

(we set N=5)K-Nearest Neighbor

(K=20)

List of Palettes

Dictionary

HQ Palettes Dictionary

LQ Palettes Dictionary

Palettes of Photo

Page 42: Introduction to my Research

Retrieved by Frequency

Retrieved by Kmeans(Cluster Center)

Proposed (Weighted Kmeans)

Finding the Dominant Colors

Page 43: Introduction to my Research

Video Demo

2012ResearchAssistant

ICPR2012

ACCV2012

>>[ Intelligent Photographing Interface with On-Device Aesthetic Quality Assessment ]

Please Visit the following link for viewing the video:

http://youtu.be/o8mKuTfO6ao

Page 44: Introduction to my Research

Discussion 1 Device : On-line assistive camera system

• Contextual Information (Viewing angle)

camera < human• Feedback

from analysis to advice• Human behavior

What do people take?How do people take?

• Computation

Server-based v.s. Device• Market and Needs

Page 45: Introduction to my Research

Discussion 2 Algorithm: photo aesthetic value assessment

• Definition of photo aesthetics

Expert v.s. Volkswagen• Labeling process

Individual bias and variance.Absolute or Relative evaluationEffect of Labeling order

• Quantify photo aesthetic

Modeling, the Personalization

Page 46: Introduction to my Research

Thanks for your attention!

10 ratings5.00/7 average

5 ratings4.90/7 average