references - university of tasmania · 2014-11-18 · references 123 popovici, v., & thiran,...

References

Bay, S.D. 1998. Combining nearest neighbor classifiers through multiple feature sub-

sets. Proc. 15th International Conference on Machine Learning, 37–45.

Belongie, Serge, Carson, Chad, Greenspan, Hayit, & Malik, Jitendra. 1998. Color-

and texture-based image segmentation using EM and its application to content-

based image retrieval. Proc. 6th International Conference on Computer Vision,

675–682.

Breiman, L. 1996. Bagging Predictors. Machine Learning, 24(2), 123–140.

Breiman, L. 1998. Arcing Classifiers. Annals of Statistics, 26(3), 801–849.

Brubaker, S.C., Wu, J., Sun, J., Mullin, M.D., & Rehg, J.M. 2005. On the Design of

Cascades of Boosted Ensembles for Face Detection. Tech. rept. GIT-GVU-05-28.

Georgia Institute of Technology.

Bryll, R., Gutierrez-Osuna, R., & Quek, F. 2003. Attribute bagging: improving accu-

racy of classifier ensembles by using random feature subsets. Pattern Recognition,

36, 1291–1302.

Chen, Y., Bart Jr, H.L., & Teng, F. 2005. A content-based image retrieval system for

fish taxonomy. Proc. 7th ACM SIGMM International Workshop on Multimedia

Information Retrieval, 237–244.

Cherkauer, K.J. 1996. Human expert-level performance on a scientific image analysis

task by a system using combined artificial neural networks. Working Notes of the

AAAI Workshop on Integrating Multiple Learned Models, 15–21.

Cohen, W.W. 1995. Fast effective rule induction. Proc. 12th International Conference

on Machine Learning, 115–123.

Dietterich, T.G. 2000. Ensemble methods in machine learning. Proc. 1st International

Workshop on Multiple Classifier Systems, 1–15.

119

REFERENCES 120

Frank, E., & Witten, I.H. 1998. Generating accurate rule sets without global opti-

mization. Proc. 15th International Conference on Machine Learning, 144–151.

Freund, Y. 1995. Boosting a weak learning algorithm by majority. Information and

Computation, 121(2), 256–285.

Freund, Y., & Schapire, R.E. 1996. Experiments with a new boosting algorithm. Proc.

13th International Conference on Machine Learning, 148–156.

Ho, T.K. 1998. The random subspace method for constructing decision forests. IEEE

Transactions on Pattern Analysis and Machine Intelligence, 20(8), 832–844.

Horton, M., Cameron-Jones, M., & Williams, R. 2006. Virtual Attribute Subsetting.

Proc. 19th Australian Joint Conference on Artificial Intelligence, 214–223.

Horton, M., Cameron-Jones, M., & Williams, R. 2007. Multiple Classifier Object

Detection with Confidence Measures. Proc. 20th Australian Joint Conference on

Artificial Intelligence, 559–568.

Hou, X., Liu, C.L., & Tan, T. 2006. Learning Boosted Asymmetric Classifiers for

Object Detection. Proc. IEEE International Conference on Computer Vision and

Pattern Recognition (CVPR’06), 330–338.

Intel. 2006. Open Source Computer Vision Library version 1.0. http: // www. intel.

com/ technology/ computing/ opencv/ (accessed 3rd August 2007).

Jacobs, C.E., Finkelstein, A., & Salesin, D.H. 1995. Fast multiresolution image query-

ing. Proc. 22nd Annual Conference on Computer Graphics and Interactive Tech-

niques, 277–286.

Jones, M., & Viola, P. 2003a. Fast multi-view face detection. Tech. rept. TR-20003-96.

Mitsubishi Electric Research Lab.

Jones, M.J., & Viola, P. 2003b. Face Recognition Using Boosted Local Features. Proc.

9th International Conference on Computer Vision.

Kearns, M.J., & Vazirani, U.V. 1994. Weak and Strong Learning in An Introduction

to Computational Learning Theory. MIT Press. Chap. 4, pages 73–102.

REFERENCES 121

Kohavi, R., Becker, B., & Sommerfield, D. 1997. Improving simple Bayes. Proc. 9th

European Conference on Machine Learning.

Kolsch, M., & Turk, M. 2004a. Analysis of Rotational Robustness of Hand Detection

with a Viola-Jones Detector. Proc. 17th International Conference on Pattern

Recognition, 107–110.

Kolsch, M., & Turk, M. 2004b. Robust hand detection. Proc. 6th IEEE International

Conference on Automatic Face and Gesture Recognition, 614–619.

Kruppa, H., Santana, M.C., & Schiele, B. 2003. Fast and robust face finding via local

context. 5th IEEE International Workshop on Visual Surveillance and Perfor-

mance Evaluation of Tracking and Surveillance, 157–164.

Kuhn, H.W. 1955. The Hungarian method for the assignment problem. Naval Re-

search Logistics Quarterly, 2, 83–97.

Langley, P. 1996. Elements of Machine Learning. Morgan Kaufmann.

Li, S.Z., & Zhang, Z. 2004. FloatBoost learning and statistical face detection. IEEE


Lienhart, R., & Maydt, J. 2002. An extended set of Haar-like features for rapid object

detection. Proc. IEEE International Conference on Image Processing (ICIP2002),

900–903.

Lienhart, R., Kuranov, A., & Pisarevsky, V. 2003a. Empirical analysis of detection

cascades of boosted classifiers for rapid object detection. DAGM 25th Pattern

Recognition Symposium, 297–304.

Lienhart, R.W., Liang, L., & Kuranov, A. 2003b. A Detector tree of boosted classifiers

for real-time object detection and tracking. Proc. IEEE International Conference

on Multimedia and Expo (ICME’03), 277–280.

Lines, JA, Tillett, RD, Ross, LG, Chan, D., Hockaday, S., & McFarlane, NJB. 2001.

An automatic image-based system for estimating the mass of free-swimming fish.

Computers and Electronics in Agriculture, 31(2), 151–168.

REFERENCES 122

Lowe, DG. 1999. Object recognition from local scale-invariant features. Proc. 7th

International Conference on Computer Vision, 1150–1157.

Luo, H. 2005. Optimization Design of Cascaded Classifiers. Proc. IEEE International

Conference on Computer Vision and Pattern Recognition, 480–485.

Messom, C., & Barczak, A. 2006. Fast and Efficient Rotated Haar-like Features

Using Rotated Integral Images. Proc. Australasian Conference on Robotics and

Automation. http: // www. araa. asn. au/ acra/ acra2006/ papers/ paper_ 5_

63. pdf .

Michie, D. 1989. Problems of computer-aided concept formation. Applications of

Expert Systems, 2, 310–333.

Mitchell, T.M. 1997. Machine Learning. McGraw-Hill.

Mueller, R.P., Brown, R.S., Hop, H., & Moulton, L. 2006. Video and acoustic camera

techniques for studying fish under ice: a review and comparison. Reviews in Fish

Biology and Fisheries, 16(2), 213–226.

Munkres, J. 1957. Algorithms for the Assignment and Transportation Problems.

Journal of the Society for Industrial and Applied Mathematics, 5(1), 32–38.

Newman, DJ, Hettich, S., Blake, CL, & Merz, CJ. 1998. UCI repository of machine

learning databases. http: // www. ics. uci. edu/ ~ mlearn/ MLRepository.

html (accessed 7th April 2006).

Oren, M., Papageorgiou, C., Sinha, P., Osuna, E., & Poggio, T. 1997. Pedestrian

detection using wavelet templates. Proc. IEEE International Conference on Com-

puter Vision and Pattern Recognition (CVPR’97), 193–199.

Papageorgiou, C., & Poggio, T. 2000. A Trainable System for Object Detection.

International Journal of Computer Vision, 38(1), 15–33.

Papageorgiou, C.P., Oren, M., & Poggio, T. 1998. A general framework for object

detection. Proc. 6th International Conference on Computer Vision, 555–562.

Pomerleau, D.A. 1989. ALVINN: an autonomous land vehicle in a neural network.

Morgan Kaufmann Publishers Inc. San Francisco, CA, USA.

REFERENCES 123

Popovici, V., & Thiran, J.P. 2003. Face Detection using an SVM Trained in Eigenfaces

space. Proc. 4th International Conference on Audio-and Video-Based Biometric

Person Authentication, 925–928.

Quinlan, JR. 1986. Induction of decision trees. Machine Learning, 1(1), 81–106.

Quinlan, J.R. 1989. Unknown attribute values in induction. Proc. 6th International

Workshop on Machine Learning, 164–168.

Quinlan, J.R. 1993. Unknown attribute values in C4.5: Programs for Machine Learn-

ing. Morgan Kaufmann. Chap. 3, page 30.

Quinlan, J.R. 1996. Bagging, Boosting, and C4.5. Proc. 13th National Conference on

Artificial Intelligence, 725–730.

Reyzin, L., & Schapire, R.E. 2006. How boosting the margin can also boost classifier

complexity. Proc. 23rd International Conference on Machine learning, 753–760.

Rowley, HA, Baluja, S., & Kanade, T. 1998a. Neural Network-Based Face Detection.

IEEE Transactions on Pattern Analysis and Machine Intelligence, 20(1), 23–38.

Rowley, HA, Baluja, S., & Kanade, T. 1998b. Rotation Invariant Neural Network-

Based Face Detection. Proc. IEEE International Conference on Computer Vision

and Pattern Recognition (CVPR’98), 38–44.

Rowley, H.A., Baluja, S., Kanade, T., Sung, K.K., & Poggio, T. 2003. Frontal Face

Images. http: // vasc. ri. cmu. edu/ idb/ html/ face/ frontal_ images/ (ac-

cessed 28th January 2008).

Sahami, M., Dumais, S., Heckerman, D., & Horvitz, E. 1998. A Bayesian approach

to filtering junk e-mail. Tech. rept. AAAI Technical Report WS-98-05. AAAI

Workshop on Learning for Text Categorization.

Schapire, R.E. 1990. The strength of weak learnability. Machine Learning, 5(2),

197–227.

Schapire, R.E., & Singer, Y. 1999. Improved Boosting Algorithms Using Confidence-

rated Predictions. Machine Learning, 37(3), 297–336.

REFERENCES 124

Schapire, R.E., Freund, Y., Bartlett, P., & Lee, W.S. 1997. Boosting the margin: A

new explanation for the effectiveness of voting methods. Proc. 14th International

Conference on Machine Learning, 322–330.

Schiele, B., & Crowley, J.L. 2000. Recognition without Correspondence using Multidi-

mensional Receptive Field Histograms. International Journal of Computer Vision,

36(1), 31–50.

Schmid, C., & Mohr, R. 1997. Local grayvalue invariants for image retrieval. IEEE


Semani, D., Saint-Jean, C., Frelicot, C., Bouwmans, T., & Courtellemont, P. 2002.

Alive Fish Species Characterization from Video Sequences. Proc. 4th International

Workshop on Statistical Techniques in Pattern Recognition, 689–698.

Sutton, C., Sindelar, M., & McCallum, A. 2005. Feature bagging: Preventing weight

undertraining in structured discriminative learning. Tech. rept. IR-402. University

of Massachusetts Center for Intelligent Information Retrieval.

Ting, K.M., & Witten, I.H. 1997. Stacked generalization: when does it work? Proc.

15th International Joint Conference on Artificial Intelligence, 866–871.

Turk, M.A., & Pentland, A. 1991. Eigenfaces for Recognition. Journal of Cognitive

Neuroscience, 3(1), 71–86.

Viola, P., & Jones, M. 2001a. Fast and Robust Classification using Asymmetric

AdaBoost and a Detector Cascade. Proc. Neural Information Processing Systems

Conference (NIPS2001), 1311–1318.

Viola, P., & Jones, M. 2001b. Rapid object detection using a boosted cascade of

simple features. Proc. IEEE International Conference on Computer Vision and

Pattern Recognition (CVPR’01), 511–518.

Viola, P., & Jones, M. 2002. Robust real-time object detection. International Journal

of Computer Vision, 57(2), 137–154.

Viola, P., Jones, M.J., & Snow, D. 2005. Detecting Pedestrians Using Patterns of

Motion and Appearance. International Journal of Computer Vision, 63(2), 153–

161.

REFERENCES 125

Weisstein, E.W. 1999. Fisher Sign Test. MathWorld–A Wolfram Web Resource. http:

// mathworld. wolfram. com/ FisherSignTest. html (accessed 6th February

2008).

Williams, R.N., Lambert, T.J., Kelsall, A.F., & Pauly, T. 2006. Detecting Marine

Animals in Underwater Video: Let’s Start with Salmon. Proc. 12th Americas

Conference on Information Systems, 1482–1490.

Witten, I.H., & Frank, E. 2005a. Cross-validation in Data Mining: Practical Machine

Learning Tools and Techniques. 2nd edn. Morgan Kaufmann. Pages 149–152.

Witten, I.H., & Frank, E. 2005b. Data Mining: Practical Machine Learning Tools

and Techniques. 2nd edn. Morgan Kaufmann.

Witten, I.H., & Frank, E. 2005c. ROC curves in Data Mining: Practical Machine

Learning Tools and Techniques. 2nd edn. Morgan Kaufmann. Pages 168–171.

Wolpert, D.H. 1992. Stacked generalization. Neural Networks, 5(2), 241–259.

Xiao, R., Zhu, L., & Zhang, H.J. 2003. Boosting chain learning for object detection.

Proc. Ninth IEEE International Conference on Computer Vision, 709–715.

Zhou, J., & Clark, C.M. 2006. Autonomous fish tracking by ROV using Monocular

Camera. Proc. 3rd Canadian Conference on Computer and Robot Vision, 68.

A Face detection tables

This appendix contains the face detection true positive and false positive counts made

by the three main methods compared in chapter 6. These counts are for detections on

the MIT/CMU frontal face testing sets A, B and C, as described in section 4.1. They

may be comparable with other detection experiments on these images, although this

thesis used its own annotations which means such comparisons won’t be exact.

Fig. 6.11 contains a graph of these counts.

126

Appendix A Face detection tables 127

Table A.1: Face detection true and false positive counts for binary detection, binary

detection followed by hill-climbing and confidence mapping

B Confidence-based ROC curves

This appendix contains the individual ROC curves plotted during rotated object de-

tection using the confidence-based methods described in chapter 6. As in section 5.4,

graphs are shown in pairs, with the first showing the angle range increasing from 0◦ to

the approximately best range, and the second showing the angle range increasing from

there to 90◦.

The curves compared in section 6.4 follow the best points from each pair of these

graphs, using the method described in section 2.7.3.1.

128

Appendix B Confidence-based ROC curves 129

0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate

Range=30°Range=25°Range=20°Range=10°Range=0°

(a) Hill-climbing, 0◦..30◦ range

0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate

Range=30°Range=45°Range=60°Range=90°

(b) Hill-climbing, 30◦..90◦ range

Figure B.1: ROC curves for fish detection on rotated images using binary detection

followed by hill-climbing, varying the cascade random angle range

0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(a) Hill-climbing, 0◦..20◦ range

0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(b) Hill-climbing, 20◦..90◦ range

Figure B.2: ROC curves for fish detection by rotated cascades using binary detection

followed by hill-climbing, varying the cascade random angle range


0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(a) Confidence mapping, no failure tolerance,

0◦..30◦ range

0

100

200

300

400

500

0 200 400 600 800

False positivesT

rue

po

sit

ive

s

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(b) Confidence mapping, no failure tolerance,

30◦..90◦ range

0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(c) Confidence mapping, failure tolerance=1,

0◦..15◦ range

0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(d) Confidence mapping, failure tolerance=1,

15◦..90◦ range

Figure B.3: ROC curves for fish detection on rotated images using confidence mapping,

varying the cascade random angle range


0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate

Range=15°

Range=10°

Range=5°

Range=0°

(a) Confidence mapping, no failure tolerance,

0◦..15◦ range

0

100

200

300

400

500

0 200 400 600 800

False positivesT

rue

po

sit

ive

s

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(b) Confidence mapping, no failure tolerance,

15◦..90◦ range

0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate

Range=15°

Range=10°

Range=5°

Range=0°

(c) Confidence mapping, failure tolerance=1

0◦..15◦ range

0

100

200

300

400

500

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(d) Confidence mapping, failure tolerance=1,

15◦..90◦ range

Figure B.4: ROC curves for fish detection by rotated cascades using confidence map-

ping, varying the cascade random angle range


0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate

Range=5°

Range=0°

(a) Seahorse heads, 0◦..5◦ range

0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(b) Seahorse heads, 5◦..90◦ range

0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate

Range=10°

Range=5°

Range=0°

(c) Seahorse bodies, 0◦..10◦ range

0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate


(d) Seahorse bodies, 10◦..90◦ range

Figure B.5: ROC curves for seahorse segment detection on rotated images using binary

detection followed by hill-climbing, varying the cascade random angle range


0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate



0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate



0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate

Range=15°

Range=10°

Range=5°

Range=0°


0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate



Figure B.6: ROC curves for seahorse segment detection on rotated images using con-

fidence mapping, varying the cascade random angle range


0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate



0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate



0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate

Range=10°

Range=5°

Range=0°


0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate



Figure B.7: ROC curves for seahorse segment detection by rotated cascades using

binary detection followed by hill-climbing, varying the cascade random angle range


0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate



0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate



0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate

Range=10°

Range=5°

Range=0°


0

200

400

600

800

0 200 400 600 800

False positives

Tru

e p

os

itiv

es

0

0.2

0.4

0.6

0.8

1

Tru

e p

os

itiv

e r

ate



Figure B.8: ROC curves for seahorse segment detection by rotated cascades using

confidence mapping, varying the cascade random angle range

C Example images with detections

This appendix shows example face, fish and seahorse images with the detections made

by the methods compared in chapter 6. The fish and seahorse segment ‘binary cascade

detections’ are also the detections made in chapter 5 when the angle step is 15◦.

136

Appendix C Example images with detections 137

(a) Original image

(b) Binary cascade detections (c) Hill-climbed detections

(d) Binary cascade detections after merging

(numbers show neighbours)

(e) Hill-climbed detections after merging


Figure C.1: Example face image with binary detections and hill-climbed detections


(a) No failure tolerance confidence map

(line lengths show ln(confidence))

(b) Failure tolerance=1 confidence map


(c) No failure tolerance confidence sum (d) Failure tolerance=1 confidence sum

(e) No failure tolerance detections

(numbers show confidence)

(f) Failure tolerance=1 detections


Figure C.2: Example face image with detections made by confidence mapping


(a) Attribute proportion=0.8 confidence map


(b) Attribute proportion=0.9 confidence map


(c) Attribute proportion=0.8 confidence sum (d) Attribute proportion=0.9 confidence sum

(e) Attribute proportion=0.8 detections


(f) Attribute proportion=0.9 detections


Figure C.3: Example face image with detections made by confidence mapping with

virtual attribute subsetting


(a) Original image






Figure C.4: Example fish image with binary detections and hill-climbed detections

made on rotated images, angle step=15◦, angle range=30◦











Figure C.5: Example fish image with detections made by confidence mapping on ro-

tated images, angle step=15◦, angle range=30◦











Figure C.6: Example fish image with detections made by confidence mapping with

virtual attribute subsetting on rotated images, angle step=15◦, angle range=30◦


(a) Original image






Figure C.7: Example fish image with binary detections and hill-climbed detections

made by rotated cascades, angle step=15◦, angle range=30◦












rotated cascades, angle step=15◦, angle range=30◦












virtual attribute subsetting and rotated cascades, angle step=15◦, angle range=30◦


Figure C.10: Original seahorse images


(a) Binary cascade detections

(b) Binary cascade detections after merging (numbers show neighbours)

Figure C.11: Example seahorse images with body (red) and head (green) detections

made on rotated images by binary cascades, angle step=15◦, head angle range=5◦,

body angle range=10◦


(a) Hill-climbed detections

(b) Hill-climbed detections after merging (numbers show neighbours)


made on rotated images by binary cascades followed by hill-climbing, angle step=15◦,

head angle range=5◦, body angle range=10◦


(a) Confidence map (line lengths show

ln(confidence)

(b) Confidence sum

(c) Detections (numbers show confidence)


made on rotated images by confidence mapping, angle step=15◦, head angle range=20◦,

body angle range=15◦


(a) Matched from binary cascade detections (fig. C.11)

(b) Matched from confidence mapped segment detections (fig. C.13)

Figure C.14: Whole seahorse detections matched from segment detections (numbers

show match cost)

references - university of tasmania · 2014-11-18 · references 123 popovici, v., & thiran,...

Documents