Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of...
Transcript of Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of...
![Page 1: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/1.jpg)
Global Probability of Boundary
Learning to Detect Natural Image Boundaries Using Local Brightness, Color, and Texture Cues – Martin, Fowlkes, Malik
Using Contours to Detect and Localize Junctions in Natural Images – Maire, Arbalaez, Fowlkes, Malik.
presented by
Varun Ramakrishna
![Page 2: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/2.jpg)
![Page 3: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/3.jpg)
![Page 4: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/4.jpg)
![Page 5: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/5.jpg)
Edge Detection Vs Boundary Detection
Edges Abrupt change in some lowlevel image feature such as brightness or color
Boundary Contour in image plane that represents a change in pixel ownership from one object to another
![Page 6: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/6.jpg)
Estimate Posterior probability of boundary passing though centre point based on local patchbased features
Using a Supervised Learning based framework
Boundary information integral to higher level tasks such as perceptual organization
Non-Boundaries Boundaries
![Page 7: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/7.jpg)
![Page 8: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/8.jpg)
Image Features
Oriented Energy
GradientBased Features
Compare contents of the
two disc halvesθr
(x,y)
![Page 9: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/9.jpg)
L*a*b* Colorspace
Which is more similar?
L*a*b* was designed to be uniform in that perceptual “closeness” corresponds to Euclidean distance in the space.
![Page 10: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/10.jpg)
L – lightness (white to black)
a – red-greeness
b – yellowness-blueness
![Page 11: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/11.jpg)
Image Features
Work in L*a*b* Colorspace – distance between points is perceptually meaningful
Kernel density estimate followed by binning
Brightness Gradient: Histogram of L* values Color Gradient : Histogram of a* b* values
![Page 12: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/12.jpg)
Image Features
Comparison of Histograms L1 Norm
Earth Mover's Distance
ChiSquared Distance
![Page 13: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/13.jpg)
Image Features
Texture Gradient 13 filter responses at each pixel Vector quantization using Kmeans Cluster centres define textons
Chisquared difference between
texton distributions
![Page 14: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/14.jpg)
Localization
Underlying function should peak at human marked boundaries
Spatially extended features On and Off boundary pixels will have a high value
![Page 15: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/15.jpg)
Localization
Improve Localization by using derived feature Divide by distance to nearest maximum
x maxima, d(x) 0, fnew large→ → →
![Page 16: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/16.jpg)
Image Features
Brightness Gradient BG(x,y,r,θ ) Color Gradient CG(x,y,r,θ ) Texture Gradient TG(x,y,r,θ )
Final set of features
{OE',BG,CG,TG'}
![Page 17: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/17.jpg)
Precision-Recall vs ROC
Framework to estimate quality of the boundary classifier
Precision: True Positives / Hypothesized Class Total
Recall: True Positives / True Class Total
![Page 18: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/18.jpg)
Recall
Precision
Class 1 Class 2
![Page 19: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/19.jpg)
Goal
FewerFalsePositives
Fewer Misses
![Page 20: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/20.jpg)
F-measure
Harmonic mean of P and R Maximum value of F along the curve Quality Measure of the P_R curve
![Page 21: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/21.jpg)
ROC Curves
![Page 22: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/22.jpg)
![Page 23: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/23.jpg)
![Page 24: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/24.jpg)
FN
TP FP
TN
Positives
Negatives
Precision = TP/(TP+FP)
Recall = TP/( TP+FN)
TPR = TP/(TP+FN)= Recall
FPR = FP/(FP+TN)
![Page 25: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/25.jpg)
FN
TP
FP
TN
Positives
Negatives
Precision = TP/(TP+FP)
Recall = TP/( TP+FN)
TPR = TP/(TP+FN)= Recall
FPR = FP/(FP+TN)
![Page 26: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/26.jpg)
![Page 27: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/27.jpg)
ROC Curves
Xaxis Fraction of false positives (fallout)
Yaxis Fraction of true positives (hit rate)
But true negatives grow as n^2, while true positives grow as n.
Fallout declines as 1/n, for a scaling of n of the image
![Page 28: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/28.jpg)
Cue-Combination• Classification Trees
– Topdown splits to maximize entropy, error bounded
• Density Estimation– Adaptive bins using kmeans
• Logistic Regression, 3 variants– Linear and quadratic terms– Confidencerated generalization of AdaBoost (Schapire&Singer)
• Hierarchical Mixtures of Experts (Jordan&Jacobs)– Up to 8 experts, initialized topdown, fit with EM
• Support Vector Machines (libsvm, Chang&Lin)– Gaussian kernel, ν parameterization
Range over bias, complexity, parametric/nonparametric
Training on 200 images from the BSDS
![Page 29: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/29.jpg)
Comparison of the different classifier models
![Page 30: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/30.jpg)
Simple logistic regression model performs as well as more complex models
Linear model supported by psychophysics (simple neuron model)
![Page 31: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/31.jpg)
Cue-Combinations
Texture gradients are an important cue!
![Page 32: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/32.jpg)
Berkeley Segmentation Dataset
Human subjects presented with image Divide into a number of segments which
represent ”things” or parts of ”things” 230 is a good number Segments should be approximately equally
important 200 images for training, 100 images for testing
![Page 33: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/33.jpg)
Canny 2MM Pb HumanImage
![Page 34: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/34.jpg)
Testing on 100 images from the BSDS
![Page 35: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/35.jpg)
Key Results
Simple model works well Texture gradient is important
.... Now combine the local cues with global cues....
![Page 36: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/36.jpg)
Line drawings convey most information Goal of boundary detection line drawings that →
would help in perceptual organization and hence object recognition
![Page 37: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/37.jpg)
Perceptual Organization
Gestaltist view of Perceptual Organization The whole is different from the sum of the
individual parts Integration of local cues as computed
previously with global cues Global Framework : Mechanism for integration
of local cues – Normalized Cuts
![Page 38: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/38.jpg)
Perceptual Organization
![Page 39: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/39.jpg)
Normalized-Cuts review
Image is modelled as a
fully connected graph
Each link between nodes
(pixels) associated with a cost
cpq measures similarity
inversely proportional
to difference in feature
![Page 40: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/40.jpg)
Find Cut that minimizes the
cost function
However large segments are penalized, so fix by normalizing for size of segments
![Page 41: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/41.jpg)
Solved by posing it as a generalized eigenvalue problem.
![Page 42: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/42.jpg)
Maximum intevening contour cue sij =Max (mPb(x,y,theta))
on line segment between pixel i and j Wij= exp(Cij/k)
Multiscale Pb
![Page 43: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/43.jpg)
Compute k+1 eigenvectors of the system arnd reshape in the size of original image – sPb
Contours extracted by taking gaussian derivatives at multiple orientations
![Page 44: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/44.jpg)
![Page 45: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/45.jpg)
The signals mPb and sPb convey different information, so a linear combination is taken and the weights are learned from training data
mPb fires at all edges sPb fires only at Salient curves
![Page 46: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/46.jpg)
![Page 47: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/47.jpg)
Experiments with LabelMe
Goal of using the boundary detection for generating line drawings that would be useful for object recognition
Interesting to see how well certain object boundaries as detected by gPb correspond with human segmentations
![Page 48: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/48.jpg)
LabelMe Dataset
Images with objects labelled gPb computed for images with certain objects Boundaries in region of object extracted from
complete image using the object mask Problems with Dataset
![Page 49: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/49.jpg)
![Page 50: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/50.jpg)
![Page 51: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/51.jpg)
![Page 52: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/52.jpg)
![Page 53: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/53.jpg)
Results
![Page 54: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/54.jpg)
![Page 55: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/55.jpg)
![Page 56: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/56.jpg)
![Page 57: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/57.jpg)
![Page 58: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/58.jpg)
![Page 59: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/59.jpg)
![Page 60: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/60.jpg)
Thank You
![Page 61: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/61.jpg)
Which is more similar?
L*a*b* was designed to be uniform in that perceptual “closeness” corresponds to Euclidean distance in the space.
![Page 62: Global Probability of Boundaryefros/courses/LBMV09/presentations/GlobalPb.pdfGlobal Probability of Boundary Learning to Detect Natural Image Boundaries Using Local Brightness, Color,](https://reader034.fdocuments.in/reader034/viewer/2022051916/6008518d9c471d19a226256a/html5/thumbnails/62.jpg)
Color