-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy path2013-04-26-Multimedia.html
686 lines (536 loc) · 19.2 KB
/
2013-04-26-Multimedia.html
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
<?xml version="1.0" encoding="utf-8"?>
<!DOCTYPE html PUBLIC "-//W3C//DTD XHTML 1.0 Strict//EN"
"http://www.w3.org/TR/xhtml1/DTD/xhtml1-strict.dtd">
<html xmlns="http://www.w3.org/1999/xhtml" lang="en" xml:lang="en">
<head>
<title>2013-04-26-Multimedia</title>
<meta http-equiv="Content-Type" content="text/html;charset=utf-8"/>
<meta name="title" content="2013-04-26-Multimedia"/>
<meta name="generator" content="Org-mode"/>
<meta name="generated" content="2013-04-26 09:55:15 PDT"/>
<meta name="author" content="Jim Blomo"/>
<meta name="description" content=""/>
<meta name="keywords" content=""/>
<link rel="stylesheet" type="text/css" href="production/common.css" />
<link rel="stylesheet" type="text/css" href="production/screen.css" media="screen" />
<link rel="stylesheet" type="text/css" href="production/projection.css" media="projection" />
<link rel="stylesheet" type="text/css" href="production/color-blue.css" media="projection" />
<link rel="stylesheet" type="text/css" href="production/presenter.css" media="presenter" />
<link href='http://fonts.googleapis.com/css?family=Lobster+Two:700|Yanone+Kaffeesatz:700|Open+Sans' rel='stylesheet' type='text/css'>
</head>
<body>
<div id="preamble">
</div>
<div id="content">
<h1 class="title">2013-04-26-Multimedia</h1>
<div id="table-of-contents">
<h2>Table of Contents</h2>
<div id="text-table-of-contents">
<ul>
<li><a href="#sec-1">1 Multimedia Data Mining</a></li>
<li><a href="#sec-2">2 Features</a></li>
<li><a href="#sec-3">3 Types</a>
<ul>
<li><a href="#sec-3-1">3.1 Covering</a></li>
</ul>
</li>
<li><a href="#sec-4">4 Generalization</a>
<ul>
<li><a href="#sec-4-1">4.1 Density</a></li>
</ul>
</li>
<li><a href="#sec-5">5 Generalized Features</a>
<ul>
<li><a href="#sec-5-1">5.1 Techniques</a></li>
</ul>
</li>
<li><a href="#sec-6">6 GIS</a></li>
<li><a href="#sec-7">7 Spatial Databases</a></li>
<li><a href="#sec-8">8 Discovery</a>
<ul>
<li><a href="#sec-8-1">8.1 Ideas</a></li>
</ul>
</li>
<li><a href="#sec-9">9 ATM Locations given obstacles</a>
<ul>
<li><a href="#sec-9-1">9.1 Yelp</a></li>
</ul>
</li>
<li><a href="#sec-10">10 Images</a>
<ul>
<li><a href="#sec-10-1">10.1 Covering</a></li>
</ul>
</li>
<li><a href="#sec-11">11 SIFT</a>
<ul>
<li><a href="#sec-11-1">11.1 Process</a></li>
</ul>
</li>
<li><a href="#sec-12">12 Sketch Recognition</a>
<ul>
<li><a href="#sec-12-1">12.1 Why?</a></li>
</ul>
</li>
<li><a href="#sec-13">13 Direction</a>
<ul>
<li><a href="#sec-13-1">13.1 Angles?</a></li>
</ul>
</li>
<li><a href="#sec-14">14 Direction Plot</a>
<ul>
<li><a href="#sec-14-1">14.1 Why?</a></li>
</ul>
</li>
<li><a href="#sec-15">15 Direction Plot</a>
<ul>
<li><a href="#sec-15-1">15.1 Why?</a></li>
</ul>
</li>
<li><a href="#sec-16">16 Features</a>
<ul>
<li><a href="#sec-16-1">16.1 Why?</a></li>
</ul>
</li>
<li><a href="#sec-17">17 All Together Now</a></li>
<li><a href="#sec-18">18 Music</a>
<ul>
<li><a href="#sec-18-1">18.1 Relation to Images</a></li>
</ul>
</li>
<li><a href="#sec-19">19 <b>Break</b></a></li>
</ul>
</div>
</div>
<div id="outline-container-1" class="outline-2">
<h2 id="sec-1"><span class="section-number-2">1</span> Multimedia Data Mining <span class="tag"><span class="slide">slide</span></span></h2>
<div class="outline-text-2" id="text-1">
</div>
</div>
<div id="outline-container-2" class="outline-2">
<h2 id="sec-2"><span class="section-number-2">2</span> Features <span class="tag"><span class="slide">slide</span> <span class="two_col">two_col</span></span></h2>
<div class="outline-text-2" id="text-2">
<ul>
<li>Core algorithms similar to "traditional" data mining
</li>
<li>Difference lies in feature engineering
</li>
<li>How to translate intuitions to numbers and formulas?
</li>
</ul>
<p> <img src="img/face-recognition.jpg" alt="img/face-recognition.jpg" />
</p>
</div>
</div>
<div id="outline-container-3" class="outline-2">
<h2 id="sec-3"><span class="section-number-2">3</span> Types <span class="tag"><span class="slide">slide</span></span></h2>
<div class="outline-text-2" id="text-3">
<dl>
<dt>Spatial</dt><dd>geographic points and features, including natural and man-made
phenomenon
</dd>
<dt>Images</dt><dd>Size, color, shape, curves, relative positions
</dd>
<dt>Music</dt><dd>Tone, tempo, beat, rhythm
</dd>
<dt>Voice</dt><dd>Speed, accent, word pauses, background noise
</dd>
</dl>
</div>
<div id="outline-container-3-1" class="outline-3">
<h3 id="sec-3-1"><span class="section-number-3">3.1</span> Covering <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-3-1">
<ul>
<li>We'll cover these areas briefly to get an overview of techniques used in
these fields
</li>
<li>All of these things <b>have</b> embedded information in them, and we are trying
to extract it
</li>
<li>One of the reasons data mining is not a black box: some one has to be on
the outside interpreting results. Results inform technique
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-4" class="outline-2">
<h2 id="sec-4"><span class="section-number-2">4</span> Generalization <span class="tag"><span class="slide">slide</span> <span class="two_col">two_col</span></span></h2>
<div class="outline-text-2" id="text-4">
<ul>
<li>Many of these areas have digital representations
</li>
<li>Can we use the raw bit representations?
</li>
<li>Usually not: must generalize patterns
</li>
</ul>
<p> <img src="img/digits.png" alt="img/digits.png" />
</p>
</div>
<div id="outline-container-4-1" class="outline-3">
<h3 id="sec-4-1"><span class="section-number-3">4.1</span> Density <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-4-1">
<ul>
<li>The data we get from digital representations is generally too sparse
</li>
<li>Key component of good learning is <b>data</b>, but you need fairly <b>dense</b> data
to learn a pattern
</li>
<li>Hypothetically, a neural network could extract general features from raw
data, but you'd need a really large amount of data in order to get the
density needed
</li>
<li>Example: for NLP, perhaps your corpus is too sparse: not many words are
shared between documents. So instead generalize: what parts of speech or
patterns show up across documents?
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-5" class="outline-2">
<h2 id="sec-5"><span class="section-number-2">5</span> Generalized Features <span class="tag"><span class="slide">slide</span></span></h2>
<div class="outline-text-2" id="text-5">
<ul>
<li>Derivative / Slope of behavior
</li>
<li>Min / Max of groups of points
</li>
<li>Bucketing / Blurring
</li>
<li>Relative positions / angles
</li>
</ul>
</div>
<div id="outline-container-5-1" class="outline-3">
<h3 id="sec-5-1"><span class="section-number-3">5.1</span> Techniques <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-5-1">
<ul>
<li>How can you strip some of the non-essential information, keep important
patterns?
</li>
<li>Many times we care about relative change, like in pricing
</li>
<li>Or group data points together (clustering is an advanced form of this)
</li>
<li>OK, let's get into some specifics:
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-6" class="outline-2">
<h2 id="sec-6"><span class="section-number-2">6</span> GIS <span class="tag"><span class="slide">slide</span> <span class="two_col">two_col</span></span></h2>
<div class="outline-text-2" id="text-6">
<ul>
<li>Geographic Information Systems
</li>
<li>Analysis and visualization of geographic data
</li>
<li>Search, terrain, object detection, flow calculations
</li>
</ul>
<p> <img src="img/gis.jpg" alt="img/gis.jpg" />
</p>
</div>
</div>
<div id="outline-container-7" class="outline-2">
<h2 id="sec-7"><span class="section-number-2">7</span> Spatial Databases <span class="tag"><span class="slide">slide</span></span></h2>
<div class="outline-text-2" id="text-7">
<ul>
<li>Integrates spatial information with traditional DBMS operations
</li>
<li>Spatial indexing, distance metrics, polygon definitions, layering
</li>
<li>Eg: Oracle Spatial Data Cartridge, ESRI Spatial Engine
</li>
</ul>
</div>
</div>
<div id="outline-container-8" class="outline-2">
<h2 id="sec-8"><span class="section-number-2">8</span> Discovery <span class="tag"><span class="slide">slide</span></span></h2>
<div class="outline-text-2" id="text-8">
<ul>
<li>What are examples of efficient city layouts?
</li>
<li>What influences successful business centers?
</li>
<li>Deforestation rates
</li>
</ul>
</div>
<div id="outline-container-8-1" class="outline-3">
<h3 id="sec-8-1"><span class="section-number-3">8.1</span> Ideas <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-8-1">
<ul>
<li>City layouts: Understanding home->work distances, not Euclidean, but
traffic on streets or by public transportation, recognizing traffic jams
</li>
<li>Business centers: analyzing network flow based on roads: industrial
supply centers nearby? Creative centers, restaurants, nightlife?
</li>
<li>Deforestation: nearby cities' effect? Recognizing forested areas vs
clear cut. Time series
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-9" class="outline-2">
<h2 id="sec-9"><span class="section-number-2">9</span> ATM Locations given obstacles <span class="tag"><span class="slide">slide</span> <span class="center">center</span></span></h2>
<div class="outline-text-2" id="text-9">
<p> <img src="img/obstacle-clustering.png" alt="img/obstacle-clustering.png" />
</p>
</div>
<div id="outline-container-9-1" class="outline-3">
<h3 id="sec-9-1"><span class="section-number-3">9.1</span> Yelp <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-9-1">
<ul>
<li>This is a current area we could improve at Yelp:
</li>
<li>Just because you're a mile from a restaurant doesn't mean it is "close"
</li>
<li>Maybe across the Bay, or maybe in between metro stops
</li>
<li>How can you calculate efficiently?
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-10" class="outline-2">
<h2 id="sec-10"><span class="section-number-2">10</span> Images <span class="tag"><span class="slide">slide</span> <span class="two_col">two_col</span></span></h2>
<div class="outline-text-2" id="text-10">
<ul>
<li>General Feature Extraction
</li>
<li>Sketch Recognition
</li>
<li>Image Recognition
</li>
</ul>
<p> <img src="img/Sift_keypoints_filtering.jpg" alt="img/Sift_keypoints_filtering.jpg" />
</p>
</div>
<div id="outline-container-10-1" class="outline-3">
<h3 id="sec-10-1"><span class="section-number-3">10.1</span> Covering <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-10-1">
<ul>
<li>We'll cover some interesting ways to extract dimensions
</li>
<li>ML/data mining combine these dimensions to do recognition with, eg.
labeled data
</li>
<li>Image on the right is using an algorithm to pick out, then filter
"interesting" points on the image
</li>
<li>img: <a href="http://en.wikipedia.org/wiki/Scale-invariant_feature_transform">http://en.wikipedia.org/wiki/Scale-invariant_feature_transform</a>
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-11" class="outline-2">
<h2 id="sec-11"><span class="section-number-2">11</span> SIFT <span class="tag"><span class="slide">slide</span></span></h2>
<div class="outline-text-2" id="text-11">
<p> <img src="img/Sift_keypoints_filtering.jpg" alt="img/Sift_keypoints_filtering.jpg" />
</p>
</div>
<div id="outline-container-11-1" class="outline-3">
<h3 id="sec-11-1"><span class="section-number-3">11.1</span> Process <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-11-1">
<ul>
<li>Successively apply Gaussian blur to image
</li>
<li>Find points which "stand out" between blurs (ie big differences)
</li>
<li>You can connect these keypoints to make a kind of fingerprint
</li>
<li>These fingerprints can be used, scaled, etc. to match against other images
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-12" class="outline-2">
<h2 id="sec-12"><span class="section-number-2">12</span> Sketch Recognition <span class="tag"><span class="slide">slide</span> <span class="center">center</span></span></h2>
<div class="outline-text-2" id="text-12">
<p> <img src="img/sketch-1.png" alt="img/sketch-1.png" />
</p><ul>
<li>Find (x,y) points along a sketch
</li>
</ul>
</div>
<div id="outline-container-12-1" class="outline-3">
<h3 id="sec-12-1"><span class="section-number-3">12.1</span> Why? <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-12-1">
<ul>
<li>Sketch recognition can be used to see if you're drawing shapes
</li>
<li>Be nice to be able to snap a picture of your diagram on a napkin and have
it come out nicely formatted?
</li>
<li>But how to recognize a circle, assuming you can't draw a perfect circle?
</li>
<li>Start with (x,y) points, but as we mentioned, very sparse
</li>
<li>Images by Marty Field
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-13" class="outline-2">
<h2 id="sec-13"><span class="section-number-2">13</span> Direction <span class="tag"><span class="slide">slide</span> <span class="center">center</span></span></h2>
<div class="outline-text-2" id="text-13">
<p> <img src="img/sketch-2.png" alt="img/sketch-2.png" />
</p><ul>
<li>Find angles along a sketch
</li>
</ul>
</div>
<div id="outline-container-13-1" class="outline-3">
<h3 id="sec-13-1"><span class="section-number-3">13.1</span> Angles? <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-13-1">
<ul>
<li>Instead of points, measure the angle at each turn
</li>
<li>You'll notice something peculiar about these angles. What?
</li>
<li>They're more than +/- 180 because we want to continue a "trend" if
they're turning the same way. Help identify changes in direction vs
spirals
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-14" class="outline-2">
<h2 id="sec-14"><span class="section-number-2">14</span> Direction Plot <span class="tag"><span class="slide">slide</span> <span class="center">center</span></span></h2>
<div class="outline-text-2" id="text-14">
<p> <img src="img/sketch-3.png" alt="img/sketch-3.png" />
</p><ul>
<li>Plot angles vs time
</li>
</ul>
</div>
<div id="outline-container-14-1" class="outline-3">
<h3 id="sec-14-1"><span class="section-number-3">14.1</span> Why? <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-14-1">
<ul>
<li>Becomes even more generalized:
<ul>
<li>What is the derivative?
</li>
<li>How many times to we change derivatives?
</li>
</ul>
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-15" class="outline-2">
<h2 id="sec-15"><span class="section-number-2">15</span> Direction Plot <span class="tag"><span class="slide">slide</span> <span class="center">center</span></span></h2>
<div class="outline-text-2" id="text-15">
<p> <img src="img/sketch-4.png" alt="img/sketch-4.png" />
</p><ul>
<li>Plot angles vs time
</li>
</ul>
</div>
<div id="outline-container-15-1" class="outline-3">
<h3 id="sec-15-1"><span class="section-number-3">15.1</span> Why? <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-15-1">
<ul>
<li>Example where we change directions
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-16" class="outline-2">
<h2 id="sec-16"><span class="section-number-2">16</span> Features <span class="tag"><span class="slide">slide</span> <span class="center">center</span></span></h2>
<div class="outline-text-2" id="text-16">
<p> <img src="img/sketch-4.png" alt="img/sketch-4.png" />
</p><dl>
<dt>NDDE</dt><dd>Normalized Distance between Direction Extremes
</dd>
<dt>DCR</dt><dd>Direction Change Ratio
</dd>
</dl>
</div>
<div id="outline-container-16-1" class="outline-3">
<h3 id="sec-16-1"><span class="section-number-3">16.1</span> Why? <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-16-1">
<dl>
<dt>NDDE</dt><dd>Are the discontinuous changes in direction, or is the line
<ul>
<li>generally curvy, and follows a similar path?
</li>
</ul>
</dd>
<dt>DCR</dt><dd>Total amount of angle change in the sketch. Low for first, high
for second
</dd>
<dt>Others?</dt><dd>bounding box size/ratio, stroke length, distance between endpoints,
length, width, height, speed, direction, acceleration
</dd>
</dl>
</div>
</div>
</div>
<div id="outline-container-17" class="outline-2">
<h2 id="sec-17"><span class="section-number-2">17</span> All Together Now <span class="tag"><span class="slide">slide</span></span></h2>
<div class="outline-text-2" id="text-17">
<iframe src="http://player.vimeo.com/video/6496886" frameborder="0" webkitAllowFullScreen mozallowfullscreen allowFullScreen></iframe> <p><a href="http://vimeo.com/6496886">Sketch2Photo: Internet Image Montage</a> from <a href="http://vimeo.com/user2276797">Tao Chen</a> on <a href="http://vimeo.com">Vimeo</a>.</p>
<p>
<a href="http://vimeo.com/6496886">Sketch2Photo</a>
</p>
</div>
</div>
<div id="outline-container-18" class="outline-2">
<h2 id="sec-18"><span class="section-number-2">18</span> Music <span class="tag"><span class="slide">slide</span></span></h2>
<div class="outline-text-2" id="text-18">
<ul>
<li>Generate a finger print: time, frequency, amplitude
</li>
<li>Filter most intense (largest) amplitudes
</li>
<li>Create a hash of connections between points
</li>
<li>Match, in time, the hash between songs
</li>
</ul>
<p> <img src="img/music_match.png" alt="img/music_match.png" />
</p>
</div>
<div id="outline-container-18-1" class="outline-3">
<h3 id="sec-18-1"><span class="section-number-3">18.1</span> Relation to Images <span class="tag"><span class="notes">notes</span></span></h3>
<div class="outline-text-3" id="text-18-1">
<ul>
<li>Interesting to note: we transformed one media type (music) into another
(image), then started using some techniques we've seen in image
fingerprinting
</li>
<li>More in reading
</li>
</ul>
</div>
</div>
</div>
<div id="outline-container-19" class="outline-2">
<h2 id="sec-19"><span class="section-number-2">19</span> <b>Break</b> <span class="tag"><span class="slide">slide</span></span></h2>
<div class="outline-text-2" id="text-19">
<script type="text/javascript" src="production/org-html-slideshow.js"></script>
</div>
</div>
</div>
<div id="postamble">
<p class="date">Date: 2013-04-26 09:55:15 PDT</p>
<p class="author">Author: Jim Blomo</p>
<p class="creator">Org version 7.8.02 with Emacs version 23</p>
<a href="http://validator.w3.org/check?uri=referer">Validate XHTML 1.0</a>
</div>
</body>
</html>