1 / 27 Detecting visi- ble areas of iris I. A. Solomatin, I. A. Matveev Detecting visible areas of iris Purpose of the by classifier of textures with support set study Problem statement Related works Proposed method Solomatin Ivan, 1. Polar transformation Matveev Ivan. 2. Feature vector. 3. Basic set 4. Training the classifier Moscow Institute of Physics and Technology 5. Classification 6. Morphological Federal Research Centre ”Computing Centre” of post-processing 7. Lacunas detection Russian Academy of Sciences Examples Experiment Step 1 October 11, 2016 Step 2 Conclusion
2 / 27 Purpose of the study Detecting visi- ble areas of iris I. A. Solomatin, I. A. Matveev Purpose: Purpose of the study To build automatic algorithm localizing visible areas of iris Problem statement using classifier, which is trained on the pixels of unoccluded Related works sector on the processed image. Proposed method 1. Polar transformation 2. Feature vector. 3. Basic set 4. Training the classifier 5. Classification 6. Morphological post-processing 7. Lacunas detection Examples Experiment Step 1 Step 2 Conclusion
3 / 27 Problem statement Detecting visi- ble areas of iris Input: I. A. Solomatin, I. A. Matveev ◮ I - grayscale bitmap sized W × H . Every pixel is Purpose of the study encoded in one byte. Problem statement ◮ x P , y P , r P - coordinates of the center and radius of the Related works circle that approximates the pupil-iris boundary. Proposed method 1. Polar ◮ x I , y I , r I - coordinates of the center and radius of the transformation 2. Feature vector. circle that approximates the sclera-iris boundary. 3. Basic set 4. Training the classifier 5. Classification 6. Morphological Output: post-processing 7. Lacunas detection Binary matrix J , sized W × H . Every pixel of the matrix Examples shows if the corresponding pixel of the source image contains Experiment Step 1 occlusion. Step 2 Conclusion J ∈ B [1; W ] × [1; H ] , where B = { 0 , 1 }
4 / 27 Problem statement Detecting visi- ble areas of iris I. A. Solomatin, In the formal way: I. A. Matveev Purpose of the Ω ⊂ [1; W ] × [1; H ]; study � � � ( x − x P ) 2 + ( y − y P ) ≥ r 2 Problem statement � � P Related works Ω = ( x , y ) : � . ( x − x I ) 2 + ( y − y I ) ≤ r 2 � Proposed method I 1. Polar transformation Ω is an annular region of the iris localization. 2. Feature vector. 3. Basic set The purpose is to classify all the pixels of Ω into two classes, 4. Training the classifier it means to build a classifier: 5. Classification 6. Morphological � post-processing 7. Lacunas detection 1 , occlusion Examples Q ( x , y ) : Ω → { 0 , 1 } . Q ( x , y ) = 0 , iris Experiment Step 1 Step 2 The results of the classification are compiled into the binary Conclusion matrix J ∈ B [1; W ] × [1; H ] , where B = { 0 , 1 } .
5 / 27 Related works Detecting visi- ble areas of iris I. A. Solomatin, I. A. Matveev Purpose of the 1. J. Daugman, How iris recognition works , ICIP (1). 2002. pp. study 33-36. Problem statement Related works In this work, boundaries of eyelids are detected using Proposed method integro-differential operators . 1. Polar transformation 2. Feature vector. 3. Basic set 2. J. Daugman, New methods in iris recognition , Systems, 4. Training the classifier 5. Classification Man, and Cybernetics, Part B, IEEE Transactions on, vol. 6. Morphological post-processing 37, no. 5, pp. 1167–1175, Oct. 2007. 7. Lacunas detection In this work, active contour approach is used to detect Examples eyelids boundaries. Experiment Step 1 Step 2 Conclusion
6 / 27 Related works Detecting visi- ble areas of iris I. A. Solomatin, 3. D. Zhang, D.M. Monro, and S. Rakshit, Eyelash removal I. A. Matveev method for human iris recognition , Image Processing, 2006 Purpose of the study IEEE International Conference on, pp. 285–288, Oct. 2006. Problem statement In this work Sobel filter is used for eyelashes detection. Related works After detection they are removed from the image using Proposed method median filter. 1. Polar transformation 2. Feature vector. 3. Basic set 4. Training the 4. Yung hui Li and Marios Savvides. A pixel-wise, classifier 5. Classification learning-based approach for occlusion estimation of iris 6. Morphological post-processing images in polar domain. In ICASSP, pages 1357-1360. IEEE, 7. Lacunas detection Examples 2009. Experiment In this work algorthm, localizing occlusions is implemented Step 1 Step 2 using classifier based on Gaussian Mixtures , which is Conclusion trained on human-made training set .
7 / 27 Proposed method. Basic stages Detecting visi- ble areas of iris 1. Applying the polar I. A. Solomatin, I. A. Matveev transformation. Purpose of the 2. Feature vector study calculation. Problem statement Related works 3. Finding the basic set S ′ . Proposed method 1. Polar transformation 4. Training the classifier, 2. Feature vector. using S ′ as training set. 3. Basic set 4. Training the classifier 5. Classification 5. Classification. 6. Morphological post-processing 7. Lacunas detection 6. Morphological Examples post-processing. Experiment Step 1 Step 2 7. Lacunas detection. Conclusion 8. Applying the reverse polar transformation.
8 / 27 Stage 1. Polar transformation Detecting visi- Coordinates of the pixels of the new region are set in ( ρ, ϕ ) ble areas of iris variables: ρ ∈ [1; h ] , ϕ ∈ [1; w ]. Every point of the polar I. A. Solomatin, I. A. Matveev domain has a prototype in the Cartesian domain according Purpose of the to the following rule: study � � � � R P + ρ 0 ( 2 πϕ Problem statement w ) ρ 2 πϕ ˆ x ( ρ, ϕ ) = x P + cos Related works h w Proposed method � � � � 1. Polar R P + ρ 0 ( 2 πϕ w ) ρ 2 πϕ transformation y ( ρ, ϕ ) = y P + ˆ sin 2. Feature vector. 3. Basic set h w 4. Training the classifier 5. Classification The Ω in the polar domain becomes a rectangle 6. Morphological post-processing Ω ′ = [1; w ] × [1; h ] 7. Lacunas detection Examples Experiment Step 1 Step 2 Conclusion
9 / 27 Stage 2. Calculating the feature vector. Detecting visi- The feature vector consists of the following K = 12 ble areas of iris components: I. A. Solomatin, I. A. Matveev ◮ B ( � x = ( ϕ, ρ ) T x ) — brightness in the point � Purpose of the ◮ B ( � x ) — average brightness in the neighbourhood of the study Problem statement point � x . Related works ◮ σ ( � x ) — the standard deviation of brightness in the Proposed method neighbourhood of � x . 1. Polar transformation ◮ � 2. Feature vector. C ( � x ) — vector of five components of discrete cosine 3. Basic set 4. Training the transform of neighbourhood of � x . DCT is calculating in classifier 5. Classification the neighbourhood sized 8 × 8. 6. Morphological post-processing ◮ � 7. Lacunas detection M ( � x ) — vector of four components of co-occurrence Examples matrix of neighbourhood of � x , which is binarized by Experiment Otsu’s threshold. Step 1 Step 2 x ∈ Ω ′ has corresponding feature vector: Every point � Conclusion x ) , ( � x )) T , ( � x )) T ) T � p ( � x ) = ( B ( � x ) , B ( � x ) , σ ( � C ( � M ( �
10 / 27 Stage 3. Basic set Detecting visi- ble areas of iris I. A. Solomatin, I. A. Matveev Then method finds unoccluded sector S , which is supposed Purpose of the to be the sector with minimum kurtosis coefficient: study Problem statement � ( X − E X ) 4 � µ = E . Related works Proposed method Fixing the central angle of the sector and solving the 1. Polar transformation problem in Ω ′ , it is possible to find sector with minimum 2. Feature vector. 3. Basic set kurtosis coefficient spending O ( wh ) time. Formal problem 4. Training the classifier statement for sector with central angle ∆ α ′ : 5. Classification 6. Morphological post-processing 4 7. Lacunas detection α ′ +∆ α ′ h α ′ +∆ α ′ h � � � � 1 Examples α ′ = argmin I ′ ( x , y ) − I ′ ( i , j ) . α ′ ∈ [0; w 2 ] ∆ α ′ h Experiment y =1 x = α ′ − ∆ α ′ i = α ′ − ∆ α ′ j =1 Step 1 Step 2 Conclusion
11 / 27 Stage 3. Basic set Detecting visi- ble areas of iris I. A. Solomatin, This figure illustrates kurtosis coefficient distribution on one I. A. Matveev of the images from CASIA database. The dark rectangle Purpose of the indicates the chosen sector. study Problem statement Related works Proposed method 1. Polar transformation 2. Feature vector. 3. Basic set 4. Training the classifier 5. Classification 6. Morphological post-processing 7. Lacunas detection Examples Experiment Step 1 Step 2 Conclusion
Recommend
More recommend
