object recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfadditional...
TRANSCRIPT
![Page 1: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/1.jpg)
1
Object Recognition
Lecture 11, April 21st, 2008
Lexing Xie
EE4830 Digital Image Processing http://www.ee.columbia.edu/~xlx/ee4830/
![Page 2: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/2.jpg)
2
Announcements
� HW#5 due today
� HW#6
� last HW of the semester
� Due May 5th
![Page 3: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/3.jpg)
3
Roadmap to Date
Spatial Domain processing and enhancement
Image Transform and Filtering
Morphological Processing
Image Descriptors
Image Segmentation
Applicatoions:Image Restoration, Object recognition, Image
Compression, Indexing and Retrieval, Reconstruction
![Page 4: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/4.jpg)
4
Lecture Outline
Problem: object recognition from images.
� What and why
� Pattern recognition primer
� Object recognition in controlled environments
� State of the art object recognition systems
![Page 5: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/5.jpg)
5
What is Object Recognition?
perceptibleperceptible materialmaterial
thingthing
Courtesy of http://people.csail.mit.edu/torralba/iccv2005/
![Page 6: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/6.jpg)
6
What is Object Recognition?
Sensory data Descriptions
Apple
+
One of the fundamental problems of computer vision:
Color, texture, shape, motion, size, weight, smell, touch, sound, …
“toy”, “stuffed Pooh”, “a frontal, close-up shot of stuffed Pooh”, “ToysRus#345812”, …
![Page 7: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/7.jpg)
7
Why?
� Science
� How do we recognize objects?
� Practice
� Robot navigation
� Medical diagnosis
� Security
� Industrial inspection and automation
� Human-computer interface
� Information retrieval
� …
![Page 8: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/8.jpg)
8
Applications of Object Recognition
Some images from http://www.cs.utexas.edu/~grauman/research/research.html
![Page 9: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/9.jpg)
9
![Page 10: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/10.jpg)
10
Lecture Outline
� Object recognition: what and why
� Object recognition in controlled environments
� Distance-based classifiers
� generalized linear classifiers
� Neural networks
� Bayes classifiers
� Object recognition in practice
� General object recognition systems
� Summary
![Page 11: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/11.jpg)
11
Objects as Vectors …
![Page 12: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/12.jpg)
12image vector representation
![Page 13: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/13.jpg)
13
pattern classifier from examples
� goal: given x, infer y
� learning from examples: supervised learning� given (xi, yi=f(xi)), i=1,…,N for some unknown function f
� find a “good approximation” to f
� rules versus data� encode human knowledge as rules
� e.g. the petal length and width of iris
� appropriate scenarios for supervised learning� no human expert (predict strength to cure AIDS given new molecule
structure)
� human can perform task but can’t describe how they do it (e.g. handwriting recognition, object recognition)
� the desired function is changing constantly w.r.t. time, or user (stock trading decisions, user-specific spam filtering)
![Page 14: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/14.jpg)
14
minimum distance classifier
step 1: calculate “class prototypes” as the means
step 2: use the prototypes to classify a new example
“discriminant” function f:
![Page 15: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/15.jpg)
15
nearest neighbor classifier
� steps:� store all training examples
� classify a new example x? by finding the training example (xi, yi) that’s nearest to x? according to Euclidean distance, and copying the labels
![Page 16: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/16.jpg)
16
nearest neighbor classifier
“discriminant” function f: gray area -1; white area +1
� (implicit) decision boundaries form a subset of the Voronoi diagram of the training data – each line segment is equidistant between two points
� comments� prone to noisy, poorly scaled features
� conditioned on the distance metric
� “smooth” the decision by looking at K-neighbors and vote
� good news: kNN is “universally consistent”
![Page 17: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/17.jpg)
17
linear classifier
� two desirables
� explicit (linear) decision boundary
� use many training examples/prototypes but do no need to remember all
![Page 18: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/18.jpg)
18
the perceptron algorithm
� learning a linear classifier
� given training data (xi, yi) and loss function
� find: weight vector [w; b] that minimizes expected loss on training data
use hinge loss:
� start from initial weights w0
� compute gradient
� update
� repeat until convergence
η : learning rate
![Page 19: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/19.jpg)
19
computing the gradient
compute gradient given
� η must decrease to zero in order to guarantee convergence.
� some algorithms (Newton’s) can automatically select η.
� local minimum is the global minimum for hinge loss
contribution from each training sample
contribution from each dimension of each training sample
![Page 20: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/20.jpg)
20
Support Vector Machines
� Two key ideas:
� The “best” separating hyperplane has the largest margin.
� Class boundary can be linear in a higher-dimensional space, e.g.,
generalized linear discriminant
weighted (generalized) inner product with “support vectors”
![Page 21: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/21.jpg)
21
Neural Networks
a single hidden layer, feed forward neural network is capable ofapproximating any continuous, multivariate function to any desired degree of accuracy and that failure to map a function arises from poor choice of network parameters, or an insufficient number of hidden neurons.
[Cybenko 1989]
![Page 22: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/22.jpg)
22
Digit Recognition with Neural Net
� LeCun et al, 1992, 1998, …http://yann.lecun.com/exdb/mnist/
0
1
![Page 23: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/23.jpg)
23
probabilistic classifiers
� what about probabilities
� p(x|y) is usually easy to obtain from training data
� can we estimate p(y|x) ?
![Page 24: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/24.jpg)
24
Bayes classifier
![Page 25: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/25.jpg)
25
Bayes classifier for Gaussian classes
![Page 26: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/26.jpg)
26
estimating the conditionals
� how do we estimate p(x|y)
� x1, x2, …, xN discrete:count over observed samplesto get the conditional histograms
�x1, x2, …, xN continuous and conditionally Gaussian
x scalar
![Page 27: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/27.jpg)
27
![Page 28: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/28.jpg)
28
Bayes classifier example
![Page 29: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/29.jpg)
29
![Page 30: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/30.jpg)
30
classification results
![Page 31: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/31.jpg)
31
homework problem 1: classifying digits
� instruction/code available
� load digits from the MNIST dataset
� baseline 1-NN classifier
� experiment/observe/improve
� k-NN, with k=3, 5
� SVM / linear classifier
� compute error rate
� examples that are correctly/incorrectly classified
![Page 32: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/32.jpg)
32
Lecture Outline
� object recognition: what and why
� object recognition as pattern classification
� general object recognition systems
� object recognition: a systems view
� current commercial systems
� real-world challenges
� survey of state-of the art
� demo websites
![Page 33: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/33.jpg)
33
Object Recognition End-to-End
Pre-processing
Feature Extraction
Sensor(s)
Classification
Post-processing
train classifier
Pre-processing
Feature Extraction
Sensor(s)
training testing
images
segments
features
classes
� window� parts or patches� salient points
� pixel vector� descriptors
� face? car?� statue of liberty?
� spatial/temporal smoothing� context, etc.
Recognition Target
![Page 34: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/34.jpg)
34
� Commercial object recognition
� Currently a $4 billion/year industry for inspection and assembly
� Almost entirely based on template matching
� Upcoming applications
� Mobile robots, toys, user interfaces
� Location recognition
� Digital camera panoramas, 3D scene modeling
Object Recognition in Practice
courtesy of David Lowe, website and CVPR 2003 Tutorial
![Page 35: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/35.jpg)
35
Industrial Applications
http://www.cs.ubc.ca/spider/lowe/vision.html
![Page 36: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/36.jpg)
36
http://www.appian-tech.com/
![Page 37: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/37.jpg)
37
http://www.sportvision.com/
![Page 38: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/38.jpg)
38
http://www.dipix.com/
![Page 39: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/39.jpg)
39
What to Recognize
Specific
Wild card Tower Bridge
Categories
building buildingbutterfly butterfly
The Stata Center
Kristen Grauman, http://www.cs.utexas.edu/~grauman/research/
![Page 40: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/40.jpg)
40
Challenges of object recognition …
![Page 41: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/41.jpg)
41
Recognize Specific Objects (1)
Appearance Matching
[Nayar, Murase et. al.]
![Page 42: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/42.jpg)
42
� PCA on the training set.
� Estimate parameters of a low-dimensional pose manifold with splines.
� Match new image to the closest point on the manifold.
![Page 43: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/43.jpg)
43
Recognize Specific Objects (2)
� Part-based approach
� Image content is transformed into local feature coordinates that are invariant to translation, rotation, scale, and other imaging parameters
� select “interest points” that are stable extrema points across different scales.
SIFT Features David Lowe, CVPR 2003 Tutorial
![Page 44: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/44.jpg)
44
SIFT Descriptor
� Thresholded image gradients are sampled over 16x16 array of locations in scale space (Gaussian-weighted).
� Create array of orientation histograms
� 8 orientations x 4x4 histogram array = 128 dimensions
David Lowe, CVPR 2003 Tutorial
![Page 45: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/45.jpg)
45
Object Category Recognition
![Page 46: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/46.jpg)
46
Overview of object category recognition …see iccv tutorial
![Page 47: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/47.jpg)
47
Demos
� Pittpatt http://demo.pittpatt.com/
![Page 48: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/48.jpg)
48It’s not just vision…
•8pm 10 Oct 05
•London
•3pm 10 Sep 05
•downloaded
from http://...
•10am 7 Sep 05
•Australian park
•Jim, Jill nearby
•4pm 8 Sep 05
•Sydney
“two koalas
seen on nat. park trip
with Jim and Jill”
“John and
his new car”
Integrate with mobile sensor information (GPS, time, nearby object or people), calendar, schedule…Infer semantically rich meta-data labels from joint sources.
“Jill and koala on
nat. park trip”
“office parking lot”
“car to consider
purchasing”
http://www.cs.utexas.edu/~grauman/research/research.html
![Page 49: Object Recognition - ee.columbia.eduxlx/courses/ee4830-sp08/notes/lect11_1up.pdfAdditional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2 nd](https://reader035.vdocuments.site/reader035/viewer/2022070217/61224c93bf4aa03d5959ea57/html5/thumbnails/49.jpg)
49
Summary
� The object recognition problem
� Pattern classification primer
� Object recognition grown up
� Readings: G&W 12.1-12.2
� Reference: Duda, Hart, Stork, “Pattern Classification”, 2nd Ed.
� Next time: Image Compression
Additional acknowledgements: Dan Ellis, EE6820 Slides; Duda, Hart& Stork, Pattern classificaion 2nd Ed., David Claus and Christoph F. Eick: Nearest Neighbor Editing and Condensing Techniques