### Create a StudySoup account

#### Be part of our community, it's free to join!

Already have a StudySoup account? Login here

# Introduction to Computer Vision CSE 185

UCMerced

GPA 3.71

### View Full Document

## 75

## 0

## Popular in Course

## Popular in Computer Science and Engineering

This 81 page Class Notes was uploaded by Abel Lueilwitz on Thursday October 29, 2015. The Class Notes belongs to CSE 185 at University of California - Merced taught by Shawn Newsam in Fall. Since its upload, it has received 75 views. For similar materials see /class/231722/cse-185-university-of-california-merced in Computer Science and Engineering at University of California - Merced.

## Reviews for Introduction to Computer Vision

### What is Karma?

#### Karma is the currency of StudySoup.

#### You can buy or earn more Karma at anytime and redeem it for class notes, study guides, flashcards, and more!

Date Created: 10/29/15

CSE185 Fall 2008 Lecture 16 Image Transformations and Spatial Filtering Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Today 0 Image transformations and spatial ltering Chap 3 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Assignments HW 3 due Wed 1029 by midnight Midterm Wed 1 15 in class 0 Lab 3 due Wed 1 112 by midnight 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Midterm Midterm on Wed 1 15 0 You will have the full 75 minutes 0 Open notesbook Bring scratch paper 0 No laptopsPDAS calculators OK 0 It will cover material through histogram equalization Lectures 116 0 Chapters 1 2 except 267 and sections 31 33 Types of questions 7 Multiple choice 7 Short answer i Computation similar to HWs 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods 5 www1mageProcessingPlacecom Of ce hour schedule next week 0 No of ce hours on Wed Nov 5 before the midterm Thurs Nov 6 I ll be out oftown Nov 4 7 199272008 R C GOnZaIeZamp R E Woods o Midterm Review Chapter 2 Elements of Visual Perception 0 Structure of the human eye 0 Light sensors on the retina rods and cones 0 Brightness adaptation and discrimination 0 Image formation geometry Light and the Electromagnetic Spectrum Image Sensing and Acquisition Image Sampling and Quantization 0 Spatial and intensity resolution 0 Image interpolation Some Basic Relationships Between Pixels 0 Neighbors of a pixel 0 Adjacency connectivity regions and boundaries 0 Distance measures An Introduction to the Mathematical Tools Used in DIP Gonzalez 5 Woods 5 www1mageProcessingPlacecom 3 Digital Image Processing 3rd ed 0 Midterm Review Chapter 2 An Introduction to the Mathematical Tools Used in DIP 0 Linear vs nonlinear operations Spatial operations Geometric spatial transformations and image registration Af ne transformations Probabilistic methods 199272008 R C GOnZaIeZamp R E Woods Gonzalez 5 Woods 5 www1mageProcessingPlacecom 3 Digital Image Processing 3rd ed 0 Midterm Review Chapter 3 Intensity transformations Image negatives Log transformations Piecewise linear transformation functions Histogram processing Histogram equalization 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1magePrOceSSingPlaceLOm w Chap 3 Image transformations and spatial filtering Histogram equalization mathematical derivation 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom 7 gt Chap 3 Image transformations and spatial filtering 0 Example 3 5 5 Before continuing it will be helpful to work through a simple example Suppose that a 3bit il nage L 8 of size 64 X 64 pixels AWN 44096 has the intensity distribution shown in Table 31 where the intensity lcvcls are in tegers in the range 0 L 1 2 0 7 rk nk 1rk ukaN 1r 7 I 790 019 r I 11123 125 r2 2 2 85 121 31 3 656 Jul r4 7 4 329 1108 r5 5 245 0116 r6 6 122 CLUB r 7quot 31 002 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods 1 www1mageProcessingplacecom 4 Chap 3 Image transformations and spatial filtering The hiatugram of our hypothetical image is sketched in Hg 319La Valult5 of the histogram equalization transformation function are obtained using Eq 338 For instance J 53 TU 725 739011 133 J39l 39 Sinlilarly l 51 Tm 12mm mm mm 308 l39 U and 53 455 5 567 3 623 35 665 55 686 57 700 This trans formation function has the staircase sh ape shown in Fig 31191 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom P m J 5 1 5k 25 o 25 39 3 i 39 20 39 39 v 39 1 O C E o I x O 39 r i i i k v u 1 2 4 5 a 7 2 3 1 r k a h c quot 1 a 3bit 8 inlensily levels image an Original FIGURE 319 quot39 o L histogram In Transformation function c Equalized histogram 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom 5 Chap 3 Image transformations and spatial filtering At this point the 539 values slill have fractions because they were generated by summing probability 1values so we round them to the nearcsl integer s 133 1 4 623 gt6 5 ans gt3 is 66597 52 455 5 56 ass gt7 s3 567 6 57 700 gt 7 These are the values of the equalized histogram Observe hat there are only ve distinct intensity levels Because rt U was mapped to 53 1 there are 790 pixels in the histogram equalized image with his value see Table 31 Also there are in this image 1023 pixels with a value of 51 3 and 850 pixels with a value of 2 5 However both 1 3 and 11 were mapped to the same value 6 so there are 656 329 t 985 pixels in lhe equalized image with this value Similarly there are 245 122 81 443 pixels with a value 01quot in the histogram equalized image Dividing these numbers by WIN 4096 yielded the equalized histogram in Fig 3196 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageprocessingPlaceLOm 5 Chap 3 Image transformations and spatial filtering at Because a histogram is an approximation to a PDF and no new allowed in lensity levels are created in the process perfectly at histograms are rare in practical applications of histogram equalization Thus unlike its continuous counterpart il cannot be proved in general that discrete histogram equaliza tion results in a uniform histogram11V However you will see shortly using Eq 333 has the general tendency to spread the histogram of the input image so than the intensity levels Of the equalized image span a wider range of the in lenrsitr scale The net result is contrast enhancement an 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 639 Woods wwwImageProcessingPlacecom Chap 3 Image transformations and spatial ltering FIGURE 320 Left column images from Fig 316 Center column corresponding histogram equalized images Right column histograms of the images in the center column 1992 2008 R C Gonzalez amp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 Histogram equalization can be generalized to histogram matching speci cation 0 Produces image With a histogram that matches a specified PDF notjust a uniform PDF 0 Section 322 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 Local histogram processing 0 So far only considered global histogram processing in Which transform is based on all pixels in image 0 Sometimes local statistics of image Will have negligible in uence on global transform 0 So instead compute transformation based on histogram in a subregion centered on the pixel to be transformed o More expensive computationally but can avoid having to re compute histogram at every location example 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 8 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering h s i Local histogram processing Section 333 abc FIGURE 326 21 Original image b Result of global histogram equalization c Result of local histogram equalization applied to a using a neighborhood of size 3 X 1992 2008 R C Gonzalez amp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom 5ng Chap 3 Image transformations and spatial filtering 0 Fundamentals of spatial ltering 0 Used for a broad spectrum of applications Enhancement Chap 3 Edge detection Chap 10 Object detection 0 The name lter is borrowed from frequency domain processing 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 Mechanics of spatial filtering 0 Spatial lter consists of 1 A neighborhood typically a small rectangle 2 A predefined operation perform on the pixels in the neighborhood 0 Filtering creates a new pixel in the output image With coordinates equal to the coordinates of the center of the neighborhood and Whose value is the result of the filtering operation 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 A processed ltered image is generated as the center of the lter Visits each pixel in the input image If the Operation performed on the pixels is linear then the lter is called a linear Spatial lter Otherwise the lter is nonlinear 199272008 R C GOnZaIeZamp R E Woods Kimage ongin y 111710 wlilJ morn mum mum warn wUJY 10111 v l In H fl 1 section under ner FIGURE 328 The mechanics of Linear spatial ltering using a 3 X 3 lter mask The form chosen to denoie the coordinates of the lter mask coefficients slmph es writing expressions for linear ltering 1992 2008 R C Gonzalez amp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering Linear spatial ltering using a 3X3 neighborhood 0 At any point xy in the image the response gxy of the filter is the sum of the products of the lter coef cients and the image pixels encompassed by the lter gxy w l lfx ly lw l0fx ly w00fxywllfxlyl Note that the center coefficient of the filter w0 0 aligns with the pixel at the location xy 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 For a mask of size m x n we assume that m2a1 and n2b1 where a and b are positive integers Thus our focus is on filters of odd size 0 In general linear spatial filtering of an image of size M x N with a filter of size m x n is given by a b gxy Z Zwstfxsyt s7a t7b where x and y are varied so that each pixel in w visits every pixel inf 0 Boundary cases where filter extends past edge of image are dealt with in a number of ways zeropadding mirroring 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 Spatial correlation and convolution 0 Are the same except in convolution the lter is rotated by l 80 degrees 199272008 R C GOnZaIeZamp R E Woods CSE185 Fall 2008 Lecture 2 Introduction continued Today Introduction continued Assignments Read Chap 1 by Wed 93 Read Chap 2 except section 267 by Wed 910 CV toplcs compress1on images Enntam 1an nfdata 7 SW m 7 mmnmcmwa One page letter 7 Sun We 4 5 nunWax I mat77 2 My Cnnsumer digital camera image 7 matsme 3by edpxxel234Wytes 7 mg77uuuupgs Satelliteimage 7 zmxnmpmx WW 1 mm 7 mg WWW CV topics compression Cam 25am 7 LnxxizuLZanrexampiz7 7 LarryGPEG arexampm Image 3251x2527p1xelsx3byvzypmd235Mbytes mad im szissmun mesmme numm mm my will minds 1 235 CV topics compression amnnm szissmun JPEGmdmmldlly 1172Gqu numm azimuzzs Iquot CV topics enhancement CV topics restoration CV topics segmentation CV topics segmentation edge detecnun CV topics representation CV topics pattern recognition 1 or i Elmf 2 fo p1ax Inn 4 939er 39 Ra 39TLJM 39Dnlk A 2 x P I XYKPS v 1 3 Newm 7 AL LAM My ALExx LawL UAJ 9 w 1 wow 94 amp Frw quotgt39 FogK Ev sznj Raga 6x6 7 y A L A E Talgb KAN Win J Aixur e in M40051 is new AIMS mm JINnn39oxs M qAaIa in d wo am guy Pun mung akua um AGEmu 0ij quot quotquotquotquot JE aha oV x DNYnAk May a ons dwewsxun 41 my a 1 gt F km MAL L Xxu j H 1 Kg Ju quot F4 WM P M Ussz psw 7 Pz Pu gtlt x1 Xs 951 PZ P L S x39 Pl y 39 39r huMaMml Brand h 031 u I r Krrrz xrrn pa X P5 YLWX39ID M PH PI 1 I I f Ps f sgts 5 F Pr I 5 j jzlj 1 P M 5 1 12 n r I Hy 5 51 P5 1 F5 39Psgt55quotjx J 5 31 30 Amman Fn epr BF hm bnmw VXp AXJ01fX 5YA 39139 SwFu i Fi 0P nu r 39 quot gt u quot 5 31 M 9 0r 3 1 Pl x x 4 A Vix AX 5300133 DC bxl39mcar surfMCI AJArA39 VLK 39 Axsahsa 25 3ng N24 p uwu HHL39QJ VSL Vina1 AJMI P uxr wjnwmwa 1CKXLquot V 11quot L 32quot Y3 10 5 SS WD i 39J p Ax 53 datum L1 Wd ux L1 Hum X m m F gt91 K2 7 xz jz I gt1 5 93 3 I LP u quotu n jq 1 W Aogtn CSE185 Fall 2008 Lecture 3 Introduction continued Madab and the Image Processing Toolbox Today Introduction continued Matlab andthe Image Processing Toolbox Assignments Read Chap 2 except section 267 by Wed 910 CV applications Multiple cameras 7 Panoramas 7 Depth from stereo 7 3D scene reconstruction a Photo Tourism 7 aka PhotoSynth at Microsott My current research projects 0nline analysis ofgeographic imagery using interest poin s di w My current research projects Imaging for monitoring air pollution httpaimowgov 75000 grant from CITRIS a Center forlnformation Technology Research in the Interest of Society a UC Berkeley Mereed Santa Cruz Davis Imaging for monitoring air pollution rm um Mint mm anew mm m miner Arrrim 1m My current research projects Using eyetracking to understand how humans view geographic imagery National Science Foundation grant a Acquisiunn nqumpmenttn Establish a Cngrritive Smsnrium and Visualizatinn Facility at UC Merced a 250000 l 1 2 it celn Kallmann Pl Teenie Madnck canl and Shawn Newsam canl My current research projects Using eyetracking to understand how humans View geographic imagery Eye aaeker inerwall Just What is an image a What does it represent a How can it be represented An image may be de ned as a 2D function f X y where x andy are the spatial coordinates and the value amplitude offat any pair of coordinates xy is called the intensity of e image at that point empty l lntvuducxmn i imam mam Imudumon What is a digital image When xy and the values offare all discrete and nite We call the image a digital image Thus a digital image is composed ofa nite number of elements each ofwhich has a particular location andv ue These elements are calledpicture elements image elements pels andor pixels chasm Madam ile human Vismnislimited m Lhevislhlepnmnn nfthe electmmagnetzc EM Spectrum images can he farmed 39nm energy m ntherpnmnns m m wmm j nunmmpmmmadm mum l E6 Immdudlun Images can also be formed by sources 7 Sound imam microscopy chapter a The ougms uf mama Image vwaaaang One sum 5st applicaqu af gtal images was m the mwspmsa mansaywhsnmsms were 5st sent by submarm cable healzen Landau and w lntxanhman afthz E aniam cable mam trmmisam system m the sany iaznsmmmauasums requredtatxanspm aplcmre aaassthz Adams mm mare Lhanaweek m isssmanz hmxs GammarRay imaging Imagmg m the Ultravmlet Band Imagmg m the szxble and In ared Bands 44 A Imagng m LheVmble and In ared Bands quot Imagmg m the szxble and In ared Bands rm 5 lm M aw c 139 a c m W Imagmg m the Visible and Infrared Bands 5 rw m 430m u Omerlmagng Madalmes Snund OLherImagmgMndalmes Snund JJJJ39 J J J Matlab programming language and so ware development environment Matlab Image Processing Toolkit IPT Matlab overview High performance language for technical computin Matlab ma1rix laboratory Anay based 7 Very linearalgebra like Made by MthWorks Matlab overview Integrates computation visualization an programming in an easy to use environment ses 7 Math and cumputauun 7 Mudelmg simulatlun and pmtutypmg 7 Data analysis and visualizauun 7 S entl c and engneenng applicahuns 7 Applicahun develupmentmcludmg graphical user interfaces Matlab overview Not cheap 7 academic prices NSSOOseat Student Version available 7 l 7 Fully funcuunal 7 m 7 Slmula un and Muaersasea Design 7 Graphical tuui fur cunnecnng mudules m awurkauw ty 3 envimnment Open source 7 Octave is a was cumputer pmgram furperfurmmg numa lcal cumputanunswhich is musdy cumpanble with MATLAB suurce Wlklpedla CSE185 Fall 2008 Lecture 17 Image Transformations and Spatial Filtering Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Today 0 Image transformations and spatial ltering Chap 3 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Assignments Midterm Wed 1 15 in class 0 Lab 3 due Wed 1 112 by midnight 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Midterm Midterm on Wed 1 15 0 You will have the full 75 minutes 0 Open notesbook Bring scratch paper 0 No laptopsPDAS calculators OK 0 It will cover material through histogram equalization Lectures 116 0 Chapters 1 2 except 267 and sections 31 33 Types of questions 7 Multiple choice 7 Short answer i Computation similar to HWs 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Of ce hour schedule this week 0 Special of ce hours Today 3 4pm Tues 1 lam noon 0 No of ce hours on Wed Nov 5 before the midterm Thurs Nov 6 I ll be out oftown Nov 5 7 199272008 R C GOnZaIeZamp R E Woods o Midterm Review Chapter 2 Elements of Visual Perception 0 Structure of the human eye 0 Light sensors on the retina rods and cones 0 Brightness adaptation and discrimination 0 Image formation geometry Light and the Electromagnetic Spectrum Image Sensing and Acquisition Image Sampling and Quantization 0 Spatial and intensity resolution 0 Image interpolation Some Basic Relationships Between Pixels 0 Neighbors of a pixel 0 Adjacency connectivity regions and boundaries 0 Distance measures An Introduction to the Mathematical Tools Used in DIP Gonzalez 5 Woods 5 www1mageProcessingPlacecom 3 Digital Image Processing 3rd ed 0 Midterm Review Chapter 2 An Introduction to the Mathematical Tools Used in DIP 0 Linear vs nonlinear operations Spatial operations Geometric spatial transformations and image registration Af ne transformations Probabilistic methods 199272008 R C GOnZaIeZamp R E Woods Gonzalez 5 Woods 5 www1mageProcessingPlacecom 3 Digital Image Processing 3rd ed 0 Midterm Review Chapter 3 Intensity transformations Image negatives Log transformations Piecewise linear transformation functions Histogram processing Histogram equalization 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 For a mask of size m x n we assume that m2a1 and n2b1 where a and b are positive integers Thus our focus is on filters of odd size 0 In general linear spatial filtering of an image of size M x N with a filter of size m x n is given by a b gxy Z Zwstfxsyt s7a t7b where x and y are varied so that each pixel in w visits every pixel inf 0 Boundary cases where filter extends past edge of image are dealt with in a number of ways zeropadding mirroring 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 Spatial correlation and convolution 0 Are the same except in convolution the lter is rotated by l 80 degrees 199272008 R C GOnZaIeZamp R E Woods Correlation Convolution Origin f w r Origin f w rotated 180 3 O O O 1 D O 0 O 1 2 3 2 8 O O 1 O O O O S 2 3 2 1 i i b D O 0 1 O 0 O O O O O 1 0 O O I j 1 2 3 2 8 8 2 3 2 l L Slarling position alignment 7 Zero padding l l c0000000100000000 DOOOOOOIODOOOOOOUc 12328 82321 d0000000100000000 00000001000000000 12328 82321 L Position after one shift 100000000 0000000100000000011 2 8 8 2 3 2 l L Position after our shiIts 60000000 23 10000000100000000 OOOUDODIUDOUUOOOUX l 2 3 2 8 S 2 3 2 1 Final position J Full Correlation result Full convolution requl g 000823210000 000123280000 0 Cropped correlation result Cropped Convolution result 11 08232100 01232800 1 FIGURE 329 quot39 39 of 1 D 39 and of a filter with 21 discrete unit impulse Note that correlation and convolution are functions of displacement 199272008 R C Gonzalez amp R E Woods f Origin fn y l l n n n n n n n u r wxv u n 1 4 u 1 2 3 u n u H r 4 5 6 u 4 u u u 7 8 9 Y Initial position for w u u 3 u 61 94 n n u u H n n n u u u H n H u n 4 1 u n n r u n 47 7 n v H II M U H 0 tr I H H u n v u I u u 4 n 6 71 u 4 I6 5 4 u u u 4 u izii u u n n H l l H H l U I 1 n u n u 1 u H M ii H U W U H U H H H n 1 n H r H II M U H p tr I H H u n v u I u u 199272008 R C Gonzalez amp R E Woods Paddcdf n I I 1 n u n n H H H H I I H 1 H l H H n H n H u H u u n ll H I n n n ll ll I H l I 8 7 1 5 4 H 2 1 u H 1 1 H H IF U u n d H I H 2 3 H 5 6 H 8 9 N u w n n u u n n u g Cropped correlation result Cropped convolution result tr H Nmn i UL130 C h u 7 l 4 1r H 1 H H u 3 H 6 1r 9 n H H convolution last row of a 2 D filter with a 2 D to simplify visual analysis Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 Correlation ofa lter wxy of size m x n with an image fxy is given by gxy wxygtlltfxy Szalatzbgwstfxsyt Convolution ofa lter wxy of size m x n with an image fxy is given by gxy wxy fxy swstfx Sy t 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering 0 For a mask of size mxn we assume that m2a1 and n2b1 where a and b are positive integers Thus our focus is on filters of odd size 0 In general linear spatial filtering of an image of size MgtltN with a filter of size mxn is given by a b gxy Z Zwstfxsyt s7a t7b where x and y are varied so that each pixel in w visits every pixel inf 0 Boundary cases where filter extends past edge of image are dealt with in a number of ways zeropadding mirroring 199272008 R C GOnZaIeZamp R E Woods gt 7 7 M ma m Wt Mkvmm mm Miami WNW Mm m dae m m pmblem No no quhe xiyammrg mawm gm l J y h ummmmmmmmm quot m gm OWMN W W 94 and cone 01 Median i Fum V 39 c I Upda ed tutonal material 1m Hfamlr39 um and xlluxmnanmn 5y mm Miami ma am mints my 1 mm ma mgtv Sim JmIMWJxlgnma ga Iiw o 01 J 1 It at Ems txwitmm lm 4mm sm hawkE mmxmmnkmm mmmmmmm ancerecogx untidemi cmionl Facial expre ionrecogni on Human pose e quotmien and racking minimum mmqmwiil ymmiw litmi 11min mt mm mmmxwwmmmg U mmwwmmw mmmma mmi ca mm dezsmandztzcmn mm mm 41mm 4 ung111mm mm a miml may Wm m l W V J we mmwmimw 1 IQ 1 1ch ehfalce bxoce mg Howui ct nbine detedion 11mm in Auxmug ikI mmm mmm 1mm at m quotm W L View ng mmmmm u39 Mmhmrub m aw unawan wmmm 39 Emma aimmmnmmwmm mmknmmmmm MW JWJ NW tecti hoe n deo I Perfo name evalumion Mm mmmwaam Mum Wm gm mm m 1 gf wm m s35 mwmmm bmmu mw gm saw n r v z mw 0mg 3E u may 6 I 39 I 33 Shh cemer p 7 mmwsasawgwwwawm Hixgtiixygt WygtZIw 1330quot mmmgxmgbmmmw w 1594quotva mwmmmmthm 3 Mama 1 Wm ml m hi milwv pi New 7 face me I De ning faces in Video I Perfonnance evalumion I Research diredion and concluding remaxks wily immm Wlmam a Emma 1 Graph Mmd mg quot a WNWNE 31W 7 irmmt Gamma m aawm vikingng W H mm fawmmw wu J hmgim mm mmmmmm m W wmmmgammm mam Imam16mg m In gaw smle mew Emawgo IN mil meme trail gt enmn Effie 1 1 l V Detggmce m39 c lorimag 5 a1 Cami non Illummanon no e Dead 1 face in Video occlu on I Perfomlance evalua on 391 ml to lewd fea ll ES in Complex I Research diredion and concluding remaka 9D back ound mvixamym xe Wm WM aimx 1 r mm Lrnta imit W193 mgwwww oWbcmlm um m Knwm W 9531wa e m Kimmy 135W e 39 solme pixel I See also Poim Dism39bmim Model PDMD mesn 1 ngi uym Stu 1mm 39 V 39 etec 1 ce 0 enumemk empla o I I Daemng face in Video nilar to knowledgeb 391 mance evalumion I Research diredion and concluding remax Appemmwg ml mummy 1 magmw wmpim f Emma Es Hamel as mum Emma quot wgnmm fmmmgivfgubuygg mm will gt FMS Furl Ag Em ammmmm m mgam 1 gig gs gww a I I M 1 61 and Poggio 94 Mam ib m wms Bampwd namzwmgnw Pm tm T mp m H l l a qua 12a 1011 t e imaginf7 ejfe ts L 1 ea in illl ni ation and different camera input gains 39 waummnm amwm nl iv mm M m m e fh i mm mm Mam MGM Mm mm neurmeun 39 xx u39 z e 1 mmle nanm nn n m u 1mm J Cami a mmm yb i 7 W mm Wig pimmq tmi m m din7 y aw ace loca on Comian a down ample he inpm image and null mil he image 2 is no 5 e e 39 n omal laces n munhim w ymgm n z wwmmwmaammj M IQ mam yjm w w m gmwgmm m Wham smawdmyma Maia Mm mm 17mm mmmmnmz mmmv irimm m A op 0R opermor o Vming Ax rmion network J qpmnmm maul mme mumm Ii dummmm Whoum aha ibaauymlivw h with W wwlmmA minmwmmm lmm w 39v a n I 1Ewugmagt m Esme sixn vimjlki wJibga We mft ii il ammnhi gamime 331 T 1116 whammy m u mewmmwmmmM ww OEOEOEO Qi w m a mum mum mm ngimw mm 3 Mn amgmmar himx mm W mulmu nqundrmic pragmmminw problem lKemelfuncho 1n nonlinear SVB 39 Fl 3 mm gmdt mmex mmm ewe W smil ibume mammmmm m e Icon In H S W m w l gasmg mmxmmwfwm 38W 9 Tinie consuminw in running and e mg Need a pick he n39gm kemel Name magma u 31 st1mg gamma vf irhyw gt mim xm mx n INN Min W39W earmrmmw n J Q u i w mimmim 1 1pmmouon r7 Class0 bm Wuxza w e w mx Auamouom a N a 05 ea Need more eciom pow 1111 feature e quotanon cl been applied 0 object recognition A m 1mm side pose a mum View memyr lwlmdrtw lvns mu m mm m mMmmmu qu wa 1 Mmmma i Am 39 Mg m WWWij m A n w n t vmmmm M m x z 7 MMquot hm WWW mmmm m ww x ganx mum mwummwgmugw mm A Mgm Ln 9 am an umzmmmm wr daemon mm Lmlmm mmmu may 9 39K a fn axmm 4 as am mi a iii w 3m imalmghm LL EUR 39S39J39W A qnmmrmmu Km m 39aimmkfmwm mm Mk mmagw 39 39 mm 39 imwmmw 5mm n w 519221 mm M wmaask wiErtan ihy Ammmwahw m A mufvt am W in mi iw Am f g i n l gm 1mm 1mm unplunzxmunn x mm t Cmmb Wmhrmwmmquot A mm mm whde an MP y m n W Th mp m b lawmm Namath E Ied V lHandle occll i0 39 Lh Implemented in Intel Open 39 Mmbmxk mmnm w r m 39igimm w amquot in dbnmmmdfngmmm lm mm mm mm cm i e m ch Need 10 we and negzmve examples Limiied v hsed approach g 4 Seenlso mm m emuem hier mamionl fenture hsedmenmd cal anal mm 1001 up able 1 Perionnance evaluation quotWm I Research direction and concluding remaiks minimum 1 W mm hxmu jmke new in m mink13quot m mw m Hm was We mmihm 39 V V V N I ectin 39 col rima weo envu onmentandllghhnv change c I Deecnn lace 1n Vldeo I Perfonnance evalua im daemon resuhs umyp 11 one line 1 I Research direc on and concluding remaka reginnsi maceBme D faim m rmmew um m i l Sa I Reduce the arch space Imma icall anug m 1 an lhp h mtr Mugmam WWW 11x mama on oNeetl m e lciem and eITec V proce the mul modal cues x 1amp4me 13mm WIth Xe lb mmgm 3911me banana r smmag mu th m ewihm mummmmmmu atin M AL v e ecu co 390 n e 39 Whm a con ecrde39 I I Deedix face in idea oDeedion ra efalsep I Perfonnance eVal 39 I R ua on Pre ion offace loca ion earch diredion and concluding remax 0 Speed hainin mm Tm v awn 4 Jmmmmw mu mm 1 m va mmu 2113 13m wgwmwmwmaw may n zu OSImag man lpr IKodnkdnm 1ananme o l mu ip epo colorimages up f e and Varying ligh ng condi ons in mailww vim was nW lm mmggmmvig agmg imam mm mm smmmwuwmm Wm 43 I 205 mg m CollededbyH Schneidennan I Some im 0 ion I De fade benchmark set What a C Receiver Wamr c a WMJH39L 111 a Whaliu umamnmw was m minimise I Different interpretation of correct detect of face location I False negat Predict a W e 1ere there I Af ct the reporting results letectioIL L actually one negative ra e I Detecting faces I Performance evaluation domair I Research dlrectlon and concluchng remarks TL mnxpwminmrm l e a r r9 s V 1 Mai in lm 4 Dylan mm mm Win 1 t m Fairgme we Mawm I ma lilile 193mg mimmam mm 3 liltion images I Preci on I Performance evaluatia CSE185 Fall 2008 Lecture 20 Image Transformations and Spatial Filtering Image Segmentation Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Today 0 Image transformations and spatial ltering Chap 3 0 Image segmentation Chap 10 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Assignments 0 Reading Chap 10 through 1026 HW 4 due Wed 1 119 by midnight 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering Sharpening spatial filters 0 We will examine various digital d erentiation techniques Fundamentally the strength of the response of a derivative filter is proportional to the degree of intensity discontinuity of the image at the point at which the operator is applied 0 Thus image differentiation enhances edges and other discontinuities such as noise and deemphasizes areas with slowly varying intensities Notes 199272008 R C GOnZaIeZamp R E Woods r Intensity transition I I l J J J 5 Constant q I intensit gt 4 i I f Ramp Step I 5 f I 3 I I 3 l E 2 I I 1 r97 Iiliii Ji Scan 7 a he 66l66gt43l21111111166l6166x lst delivative 0 l 1 l 1 l O 0 O O O 5 0 0 0 2nd delivative O O l 0 0 0 0 I 0 O O 0 5 5 0 O 0 5 IE 4 3 2 2 li Yquot i I gt 3 1 EL I F39 r I V g 0 m EhEk El ll llll J a E E p E E E d 1 quot39 l A s lquot 2310 messing quot l39 72 39l I 3 o Filst derivative 139 4 3 Second derivative 5 5 199272008 R C GOnZaIeZamp R E Woods a b c FIGURE 336 Illustration of the rst and second derivatives of 3 representing a section of a horizontal intensity pro le from an image In a and C data points are joined by clashed lines as a visualization aid Digital Image Processing 3rd ed Gonzalez 5 Woods wwwImageProcessingPlacecom Chap 3 Image transformations and spatial ltering Different lter masks used to implement Laplacian What is the motivation behind b and 1 1992 2008 R C Gonzalez amp R E Woods 11 1 o 1 1 1 1 74 1 1 is 1 11 1 1 1 1 1 u 1 0 71 71 1 71 4 71 71 8 71 0 a1 0 71 71 1 ab Cd FIGURE 337 a Filter mask used l0 implement Eq 366 b Mask used to implement an extension of this equation that includes the diagonal terms c and 1 Two other implementa tions of the Laplacian found frequently in practice 1992 2008 R C Gonzalez amp R E Woods a b c d 6 FIGURE 338 a Blurred image of the North Pole 0 I e moo scaling d Image sharpened using the mask in 3 3373 e Result in Fig 3371 Original imag courtesy of NASA CSE185 Fall 2008 Lecture 1 Introduction Today 0 Course specifics 0 Introduction Assignments 0 Read Chap 1 by Wed 93 Computer Vision ltgt Computer Graphics computer Vision computer graphics Image Processing ltgt Computer Vision image processing computer vision acquisition compression enhancement restoration segmentation representation features pattern recognition understanding Image Processing and Computer Vision analysis in the spatial domain analysis in the frequency domain Fourier analysis etc CSE185 Fall 2008 Lecture 21 Image Transformations and Spatial Filtering Image Segmentation Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Today 0 Image transformations and spatial ltering Chap 3 0 Image segmentation Chap 10 199272008 R C GOnZaIeZamp R E Woods Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageprocessingplacecom Assignments HW 4 due Wed 1 119 by midnight 0 Reading Chap 10 through 1042 by 121 Lab 5 due Wed 123 by midnight Digital Image Processing 3rd ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering Sharpening spatial filters 0 We will examine various digital d erentiation techniques Fundamentally the strength of the response of a derivative filter is proportional to the degree of intensity discontinuity of the image at the point at which the operator is applied 0 Thus image differentiation enhances edges and other discontinuities such as noise and deemphasizes areas with slowly varying intensities Notes 199272008 R C GOnZaIeZamp R E Woods r Intensity transition I I l J J J 5 Constant q I intensit gt 4 i I f Ramp Step I 5 f I 3 I I 3 l E 2 I I 1 r97 Iiliii Ji Scan 7 a he 66l66gt43l21111111166l6166x lst delivative 0 l 1 l 1 l O 0 O O O 5 0 0 0 2nd delivative O O l 0 0 0 0 I 0 O O 0 5 5 0 O 0 5 IE 4 3 2 2 li Yquot i I gt 3 1 EL I F39 r I V g 0 m EhEk El ll llll J a E E p E E E d 1 quot39 l A s lquot 2310 messing quot l39 72 39l I 3 o Filst derivative 139 4 3 Second derivative 5 5 199272008 R C GOnZaIeZamp R E Woods a b c FIGURE 336 Illustration of the rst and second derivatives of 3 representing a section of a horizontal intensity pro le from an image In a and C data points are joined by clashed lines as a visualization aid Digital Image Processing 3rd ed Gonzalez 5 Woods wwwImageProcessingPlacecom Chap 3 Image transformations and spatial ltering Different lter masks used to implement Laplacian What is the motivation behind b and 1 1992 2008 R C Gonzalez amp R E Woods 11 1 o 1 1 1 1 74 1 1 is 1 11 1 1 1 1 1 u 1 0 71 71 1 71 4 71 71 8 71 0 a1 0 71 71 1 ab Cd FIGURE 337 a Filter mask used l0 implement Eq 366 b Mask used to implement an extension of this equation that includes the diagonal terms c and 1 Two other implementa tions of the Laplacian found frequently in practice 1992 2008 R C Gonzalez amp R E Woods a b c d 6 FIGURE 338 a Blurred image of the North Pole 0 I e moo scaling d Image sharpened using the mask in 3 3373 e Result in Fig 3371 Original imag courtesy of NASA Digital Image Processing 31d ed Gonzalez 5 Woods www1mageProcessingPlacecom Chap 3 Image transformations and spatial filtering Original signal 20 UN FIGURE 339 l D illustration of the mechanics of unsharp inaskjng a Original signal b Blurred signal Wit original shown dashed for refere ncc C Insharp mask d Sharp euecl signal obtained by H adding C to a Unslmrp mask sharpened signal 1992 2008 R C Gonzalez amp R E Woods Digital Image Processing 31d ed Gonzalez 8 Woods www1mageProcessingPIacecom Chap 3 Image transformations and spatial filtering DIPXE 3 d 6 FIGURE 340 DIP XE image b Result of blurring With a G3 ssian filter 5 DIP XE EB I 23 5 I quotO E 1992 2008 R C Gonzalez amp R E Woods

### BOOM! Enjoy Your Free Notes!

We've added these Notes to your profile, click here to view them now.

### You're already Subscribed!

Looks like you've already subscribed to StudySoup, you won't need to purchase another subscription to get this material. To access this material simply click 'View Full Document'

## Why people love StudySoup

#### "There's no way I would have passed my Organic Chemistry class this semester without the notes and study guides I got from StudySoup."

#### "When you're taking detailed notes and trying to help everyone else out in the class, it really helps you learn and understand the material...plus I made $280 on my first study guide!"

#### "I was shooting for a perfect 4.0 GPA this semester. Having StudySoup as a study aid was critical to helping me achieve my goal...and I nailed it!"

#### "Their 'Elite Notetakers' are making over $1,200/month in sales by creating high quality content that helps their classmates in a time of need."

### Refund Policy

#### STUDYSOUP CANCELLATION POLICY

All subscriptions to StudySoup are paid in full at the time of subscribing. To change your credit card information or to cancel your subscription, go to "Edit Settings". All credit card information will be available there. If you should decide to cancel your subscription, it will continue to be valid until the next payment period, as all payments for the current period were made in advance. For special circumstances, please email support@studysoup.com

#### STUDYSOUP REFUND POLICY

StudySoup has more than 1 million course-specific study resources to help students study smarter. If you’re having trouble finding what you’re looking for, our customer support team can help you find what you need! Feel free to contact them here: support@studysoup.com

Recurring Subscriptions: If you have canceled your recurring subscription on the day of renewal and have not downloaded any documents, you may request a refund by submitting an email to support@studysoup.com

Satisfaction Guarantee: If you’re not satisfied with your subscription, you can contact us for further help. Contact must be made within 3 business days of your subscription purchase and your refund request will be subject for review.

Please Note: Refunds can never be provided more than 30 days after the initial purchase date regardless of your activity on the site.