Student: Chih-Wei Fang ( 方志偉 ) Adviser: Jenn-Jier James Lien ( 連震杰 ) Robotics Laboratory, Department of Computer Science and Information Engineering, National.

Slides:



Advertisements
Similar presentations
Advanced Image Processing Student Seminar: Lipreading Method using color extraction method and eigenspace technique ( Yasuyuki Nakata and Moritoshi Ando.
Advertisements

Object Removal by Exemplar-Based Inpainting Ye Hong CS766 Fall 2004.
Image Repairing: Robust Image Synthesis by Adaptive ND Tensor Voting IEEE Computer Society Conference on Computer Vision and Pattern Recognition Jiaya.
Image and Video Upscaling from Local Self Examples
Face Recognition and Biometric Systems Eigenfaces (2)
Principal Component Analysis Based on L1-Norm Maximization Nojun Kwak IEEE Transactions on Pattern Analysis and Machine Intelligence, 2008.
1.  Texturing is a core process for modeling surface details in computer graphics applications › Texture mapping › Surface texture synthesis › Procedural.
Instructor: Mircea Nicolescu Lecture 15 CS 485 / 685 Computer Vision.
Image Denoising using Locally Learned Dictionaries Priyam Chatterjee Peyman Milanfar Dept. of Electrical Engineering University of California, Santa Cruz.
EI San Jose, CA Slide No. 1 Measurement of Ringing Artifacts in JPEG Images* Xiaojun Feng Jan P. Allebach Purdue University - West Lafayette, IN.
Texture Segmentation Based on Voting of Blocks, Bayesian Flooding and Region Merging C. Panagiotakis (1), I. Grinias (2) and G. Tziritas (3)
Effective Image Database Search via Dimensionality Reduction Anders Bjorholm Dahl and Henrik Aanæs IEEE Computer Society Conference on Computer Vision.
1 Image Completion using Global Optimization Presented by Tingfan Wu.
Region Segmentation. Find sets of pixels, such that All pixels in region i satisfy some constraint of similarity.
A Study of Approaches for Object Recognition
Segmentation Divide the image into segments. Each segment:
Detecting Image Region Duplication Using SIFT Features March 16, ICASSP 2010 Dallas, TX Xunyu Pan and Siwei Lyu Computer Science Department University.
Fast Image Replacement Using Multi-Resolution Approach Chih-Wei Fang and Jenn-Jier James Lien Robotics Lab Department of Computer Science and Information.
Texture Synthesis on Surfaces Paper by Greg Turk Presentation by Jon Super.
Face Recognition Using Eigenfaces
Region Filling and Object Removal by Exemplar-Based Image Inpainting
Smart Traveller with Visual Translator for OCR and Face Recognition LYU0203 FYP.
Multiple Object Class Detection with a Generative Model K. Mikolajczyk, B. Leibe and B. Schiele Carolina Galleguillos.
Face Processing System Presented by: Harvest Jang Group meeting Fall 2002.
Chrominance edge preserving grayscale transformation with approximate first principal component for color edge detection Professor: 連震杰 教授 Reporter: 第17組.
Computer Vision - A Modern Approach Set: Segmentation Slides by D.A. Forsyth Segmentation and Grouping Motivation: not information is evidence Obtain a.
Background Estimation Mehdi Ghayoumi, MD Iftakharul Islam, Muslem Al-Saidi Department of Computer Science Kent State University, Kent, OH
CS 485/685 Computer Vision Face Recognition Using Principal Components Analysis (PCA) M. Turk, A. Pentland, "Eigenfaces for Recognition", Journal of Cognitive.
Multiclass object recognition
CSCE 441: Computer Graphics Image Filtering Jinxiang Chai.
Distinctive Image Features from Scale-Invariant Keypoints By David G. Lowe, University of British Columbia Presented by: Tim Havinga, Joël van Neerbos.
Computer vision.
Training Database Step 1 : In general approach of PCA, each image is divided into nxn blocks or pixels. Then all pixel values are taken into a single one.
Introduction to Visible Watermarking IPR Course: TA Lecture 2002/12/18 NTU CSIE R105.
Database-Assisted Low-Dose CT Image Restoration Klaus Mueller Computer Science Lab for Visual Analytics and Imaging (VAI) Stony Brook University Wei Xu,
Texture Optimization for Example-based Synthesis Vivek Kwatra Irfan Essa Aaron Bobick Nipun Kwatra.
Digital Image Processing CCS331 Relationships of Pixel 1.
Previous lecture Texture Synthesis Texture Transfer + =
Computer Vision Lab. SNU Young Ki Baik Nonlinear Dimensionality Reduction Approach (ISOMAP, LLE)
Structured Face Hallucination Chih-Yuan Yang Sifei Liu Ming-Hsuan Yang Electrical Engineering and Computer Science 1.
TEXTURE SYNTHESIS BY NON-PARAMETRIC SAMPLING VIVA-VITAL Nazia Tabassum 27 July 2015.
Chin-Hsien Fang( 方競賢 ), Ju-Chin Chen( 陳洳瑾 ), Chien-Chung Tseng( 曾建中 ),and Jenn-Jier James Lien( 連震杰 ) Department of Computer Science and Information Engineering,
Computer Vision Lecture #10 Hossam Abdelmunim 1 & Aly A. Farag 2 1 Computer & Systems Engineering Department, Ain Shams University, Cairo, Egypt 2 Electerical.
Digital imaging By : Alanoud Al Saleh. History: It started in 1960 by the National Aeronautics and Space Administration (NASA). The technology of digital.
Design of PCA and SVM based face recognition system for intelligent robots Department of Electrical Engineering, Southern Taiwan University, Tainan County,
2D Texture Synthesis Instructor: Yizhou Yu. Texture synthesis Goal: increase texture resolution yet keep local texture variation.
Digital imaging By : Alanoud Al Saleh. History: It started in 1960 by the National Aeronautics and Space Administration (NASA). The technology of digital.
Supervisor: Nakhmani Arie Semester: Winter 2007 Target Recognition Harmatz Isca.
A Multiresolution Symbolic Representation of Time Series Vasileios Megalooikonomou Qiang Wang Guo Li Christos Faloutsos Presented by Rui Li.
MIT AI Lab / LIDS Laboatory for Information and Decision Systems & Artificial Intelligence Laboratory Massachusetts Institute of Technology A Unified Multiresolution.
Geometry Synthesis Ares Lagae Olivier Dumont Philip Dutré Department of Computer Science Katholieke Universiteit Leuven 10 August, 2004.
研 究 生:周暘庭 Q36994477 電腦與通信工程研究所 通訊與網路組 指導教授 :楊家輝 Mean-Shift-Based Color Tracking in Illuminance Change.
Irfan Ullah Department of Information and Communication Engineering Myongji university, Yongin, South Korea Copyright © solarlits.com.
Speaker Min-Koo Kang March 26, 2013 Depth Enhancement Technique by Sensor Fusion: MRF-based approach.
Statistical Models of Appearance for Computer Vision 主講人:虞台文.
Shadow Detection in Remotely Sensed Images Based on Self-Adaptive Feature Selection Jiahang Liu, Tao Fang, and Deren Li IEEE TRANSACTIONS ON GEOSCIENCE.
- photometric aspects of image formation gray level images
CS262: Computer Vision Lect 09: SIFT Descriptors
Project 1: hybrid images
Unsupervised Riemannian Clustering of Probability Density Functions
Palm Oil Plantation Area Clusterization for Monitoring
Feature description and matching
Cheng-Ming Huang, Wen-Hung Liao Department of Computer Science
By Pradeep C.Venkat Srinath Srinivasan
Data-driven methods: Texture 2 (Sz 10.5)
Dingding Liu* Yingen Xiong† Linda Shapiro* Kari Pulli†
Face Recognition and Detection Using Eigenfaces
Outline S. C. Zhu, X. Liu, and Y. Wu, “Exploring Texture Ensembles by Efficient Markov Chain Monte Carlo”, IEEE Transactions On Pattern Analysis And Machine.
Local Binary Patterns (LBP)
Blobworld Texture Features
Presentation transcript:

Student: Chih-Wei Fang ( 方志偉 ) Adviser: Jenn-Jier James Lien ( 連震杰 ) Robotics Laboratory, Department of Computer Science and Information Engineering, National Cheng Kung University, Tainan, Taiwan, R.O.C. Texture Analysis and Synthesis Using Directional Patch-Based Sampling

Motivation 2

Related Work Onion-peel approach [5] A. Criminisi, P. Perez, and K. Toyama, “Region Filling and Object Removal by Exemplar-Based Image Inpainting,” IEEE Trans. on Image Processing, Vol. 13, No. 9, pp ,

Related Work [5] A. Criminisi, P. Perez, and K. Toyama, “Region Filling and Object Removal by Exemplar-Based Image Inpainting,” IEEE Trans. on Image Processing, Vol. 13, No. 9, pp , P(p) = C(p)D(p) Confidence term Data term 4

Problems of Ref [5] Larger removal region is not easy to converge. Influenced by high-frequency components or noise. Filling in the matching patch from searching exhaustively. 5

System Framework – Multi-resolution Down-sampling for Training The training process extending from the input image to get the weight vector of each patch. Up-sampling for Image Completion Directional image completion to propagate the structure. Non-directional image completion to maintain detailed features. 6

Down-sampling from levels 0 to L I 1 I 2 I 3 I L=4 α 1 α 2 α 3 α L=4 Input image I 0 Inverse matte α 0 Down-sampling for Training Total M patches in background regions from levels 1 to L … … Select O-shaped patterns for training … … Create eigenspace Ψ Project the M O-shaped patterns onto the first N eigenvectors to have corresponding weight vectors. Cluster weight vectors using VQ. 7

Sampling It is unnecessary to use entire patches of image I 0. It will include many unnecessary patches. Require a large amount of template matching operations. Image I 0 contains more noises affect the result of the PCA process. O-shaped pattern insteads of whole patch for the training data. Whole patch elements to the training data for further matching comparison may result in a discontinuous structure of the patch. Increase the training time certainly. Search PatchO-shaped Pattern O-shaped Pattern Vector Wp Hp ω K pixels … 8

PCA Reduce the dimensions of the data representation 9

Recombine the features of the O-shaped pattern First several eigenvectors => control the global geometrical structure. The middle eigenvectors => control the local detailed features. The last few eigenvectors => control some noises. Eigenvectors 10

Projection and VQ Projecting O-shaped Patterns onto Eigenspace Ψ Reducing dimensions from K to N Vector quantization (VQ) is adopted for clustering the projection weight vectors in the eigenspace so as to reduce the comparison time. 11

System Framework Down-sampling for Training The training process extending from the input image to get the weight vector of each patch. Up-sampling for Image Completion Directional image completion to propagate the structure. Non-directional image completion to maintain detailed features. 12

Level L-1: Directional and Non-Directional IC 1. Starting from patch having larger Hessian matrix decision value (HMDV) 2. If (HMDV) ≧ Threshold: Then Search along the 1st eigenvector v 1 direction of Hessian matrix (directional). Else Texture synthesis (non-directional). Up-sampling for Image Completion Level L: Initialize synthesis values. Levels L-2 to 1: Non-Directional IC 1. Starting from patch having larger HMDV. 2. Texture synthesis. Level 0: Texture Refinement 1. Starting from patch having larger HMDV. 2. Texture synthesis. 3. Find the best matching patch from image I 0. FiFi BiBi v1v1 λ 1 I0I0 I i, i=1~L FiFi BiBi Decision window 13

Filling Order Hessian Matrix => Obtain λ 1 and λ 2 (λ 1 ≧ λ 2 ) Hessian Matrix Decision Value 14

Hessian Matrix Decision Value Higher HMDV value V (V >> 1.0 or λ 1 >> λ 2 ) directional and exists stronger edge along eigenvector v 1. If the HMDV value V is close to or equal to 1 If both λ 1 and λ 2 have higher values More detailed features or high-frequency noises. If both λ 1 and λ 2 have lower values Smooth patch. 15

Hessian Matrix Decision Value 16

Hessian Matrix Decision Value 17

Level L-1: Directional and Non-Directional IC 1. Starting from patch having larger Hessian matrix decision value (HMDV) 2. If (HMDV) ≧ Threshold: Then Search along the 1st eigenvector v 1 direction of Hessian matrix (directional). Else Texture synthesis (non-directional). Up-sampling for Image Completion Level L: Initialize synthesis values. Levels L-2 to 1: Non-Directional IC 1. Starting from patch having larger HMDV. 2. Texture synthesis. Level 0: Texture Refinement 1. Starting from patch having larger HMDV. 2. Texture synthesis. 3. Find the best matching patch from image I 0. FiFi BiBi v1v1 λ 1 I0I0 I i, i=1~L FiFi BiBi Decision window 18

Level L-1: Directional and Non-Directional IC 1. Starting from patch having larger Hessian matrix decision value (HMDV) 2. If (HMDV) ≧ Threshold: Then Search along the 1st eigenvector v 1 direction of Hessian matrix (directional). Else Texture synthesis (non-directional). Up-sampling for Image Completion Level L: Initialize synthesis values. Levels L-2 to 1: Non-Directional IC 1. Starting from patch having larger HMDV. 2. Texture synthesis. Level 0: Texture Refinement 1. Starting from patch having larger HMDV. 2. Texture synthesis. 3. Find the best matching patch from image I 0. FiFi BiBi v1v1 λ 1 I0I0 I i, i=1~L FiFi BiBi Decision window 19

Texture Synthesis The O-shaped pattern of each search patch is projected onto the eigenspace Ψ to obtain the corresponding weight vector. Based on the similarity measure of Euclidean distance between the weight vector of this search pattern and those of the cluster centers, this search pattern will be classified to the nearest cluster. Then this search pattern will compare with all patterns within the same cluster to find the best matching pattern. The patch corresponding to the best matching pattern will be directly pasted onto the location of the search patch. 20

Level L-1: Directional and Non-Directional IC 1. Starting from patch having larger Hessian matrix decision value (HMDV) 2. If (HMDV) ≧ Threshold: Then Search along the 1st eigenvector v 1 direction of Hessian matrix (directional). Else Texture synthesis (non-directional). Up-sampling for Image Completion Level L: Initialize synthesis values. Levels L-2 to 1: Non-Directional IC 1. Starting from patch having larger HMDV. 2. Texture synthesis. Level 0: Texture Refinement 1. Starting from patch having larger HMDV. 2. Texture synthesis. 3. Find the best matching patch from image I 0. FiFi BiBi v1v1 λ 1 I0I0 I i, i=1~L FiFi BiBi Decision window 21

Experimental Results Image Inpainting Bertalmio et al. [2] ACM SIGGRAPH Exemplar-Based Image Inpainting Criminisi et al. [5] IEEE Trans. on Image Processing Our Method Input Image C4C4 C3C3 C2C2 C1C1 C0C0 22

Experimental Results Image Inpainting Bertalmio et al. [2] ACM SIGGRAPH Exemplar-Based Image Inpainting Criminisi et al. [5] IEEE Trans. on Image Processing Our MethodInput Image 23

Experimental Results Our MethodInput Image 24 Exemplar-Based Image Inpainting Criminisi et al. [5] IEEE Trans. on Image Processing Image Inpainting Bertalmio et al. [2] ACM SIGGRAPH

Experimental Results Processing Time Our Method Exemplar-Based Image Inpainting Criminisi et al. [5] Image Inpainting Bertalmio et al. [2] Windmill11 Second139 Second2 Second Diving2 Second50 Second1 Second Slope4 Second61 Second1 Second 25

Conclusions Multi-Resolution Accelerate the convergence of the system and Handle the large removed region More training patches with various scales Hessian Matrix Decision Directional image completion to propagate the structure Non-directional image completion to maintain detailed features. Training for input image Reduce the computational time during the similarity measure at the synthesis process. 26

Thanks! 27