Speaker Min-Koo Kang November 14, 2012 Depth Enhancement Technique by Sensor Fusion: Joint Bilateral Filter Approaches.

Slides:



Advertisements
Similar presentations
Bayesian Belief Propagation
Advertisements

Object Recognition from Local Scale-Invariant Features David G. Lowe Presented by Ashley L. Kapron.
CSCE 643 Computer Vision: Template Matching, Image Pyramids and Denoising Jinxiang Chai.
Hongliang Li, Senior Member, IEEE, Linfeng Xu, Member, IEEE, and Guanghui Liu Face Hallucination via Similarity Constraints.
Accelerating Spatially Varying Gaussian Filters Jongmin Baek and David E. Jacobs Stanford University.
M.S. Student, Hee-Jong Hong
Real-Time Accurate Stereo Matching using Modified Two-Pass Aggregation and Winner- Take-All Guided Dynamic Programming Xuefeng Chang, Zhong Zhou, Yingjie.
IMAGE UPSAMPLING VIA IMPOSED EDGE STATISTICS Raanan Fattal. ACM Siggraph 2007 Presenter: 이성호.
December 5, 2013Computer Vision Lecture 20: Hidden Markov Models/Depth 1 Stereo Vision Due to the limited resolution of images, increasing the baseline.
Patch Based Synthesis for Single Depth Image Super-Resolution (ECCV 2012) Oisin Mac Aodha, Neill Campbell, Arun Nair and Gabriel J. Brostow Presented By:
Depth Coding Using a Boundary Reconstruction Filter for 3-D Video Systems Kwan-Jung Oh, Anthony Vetro, Fellow, IEEE, Yo-Sung Ho, Senior Member, IEEE CSVT,
Yen-Lin Lee and Truong Nguyen ECE Dept., UCSD, La Jolla, CA Method and Architecture Design for Motion Compensated Frame Interpolation in High-Definition.
Boundary matting for view synthesis Samuel W. Hasinoff Sing Bing Kang Richard Szeliski Computer Vision and Image Understanding 103 (2006) 22–32.
Technion - Israel Institute of Technology 1 Interpolation Method using Statistical Models RONEN SHER Supervisor: MOSHE PORAT.
Probabilistic video stabilization using Kalman filtering and mosaicking.
CSCE 641 Computer Graphics: Image-based Rendering (cont.) Jinxiang Chai.
Stereo Computation using Iterative Graph-Cuts
Scale-Invariant Feature Transform (SIFT) Jinxiang Chai.
Stereo matching Class 10 Read Chapter 7 Tsukuba dataset.
Face Recognition Using Neural Networks Presented By: Hadis Mohseni Leila Taghavi Atefeh Mirsafian.
Super-Resolution of Remotely-Sensed Images Using a Learning-Based Approach Isabelle Bégin and Frank P. Ferrie Abstract Super-resolution addresses the problem.
A Gentle Introduction to Bilateral Filtering and its Applications Limitation? Pierre Kornprobst (INRIA) 0:20.
CSCE 441: Computer Graphics Image Filtering Jinxiang Chai.
Fast Bilateral Filtering
A Gentle Introduction to Bilateral Filtering and its Applications 08/10: Applications: Advanced uses of Bilateral Filters Jack Tumblin – EECS, Northwestern.
Stereo Matching Information Permeability For Stereo Matching – Cevahir Cigla and A.Aydın Alatan – Signal Processing: Image Communication, 2013 Radiometric.
ICPR/WDIA-2012 High Quality Novel View Synthesis Based on Low Resolution Depth Image and High Resolution Color Image Jui-Chiu Chiang, Zheng-Feng Liu, and.
Automatic Registration of Color Images to 3D Geometry Computer Graphics International 2009 Yunzhen Li and Kok-Lim Low School of Computing National University.
Introduction to Visible Watermarking IPR Course: TA Lecture 2002/12/18 NTU CSIE R105.
A Local Adaptive Approach for Dense Stereo Matching in Architectural Scene Reconstruction C. Stentoumis 1, L. Grammatikopoulos 2, I. Kalisperakis 2, E.
Takuya Matsuo, Norishige Fukushima and Yutaka Ishibashi
High-Resolution Interactive Panoramas with MPEG-4 발표자 : 김영백 임베디드시스템연구실.
Robust global motion estimation and novel updating strategy for sprite generation IET Image Processing, Mar H.K. Cheung and W.C. Siu The Hong Kong.
December 4, 2014Computer Vision Lecture 22: Depth 1 Stereo Vision Comparing the similar triangles PMC l and p l LC l, we get: Similarly, for PNC r and.
INFORMATIK Mesh Smoothing by Adaptive and Anisotropic Gaussian Filter Applied to Mesh Normals Max-Planck-Institut für Informatik Saarbrücken, Germany Yutaka.
Image Processing Edge detection Filtering: Noise suppresion.
03/05/03© 2003 University of Wisconsin Last Time Tone Reproduction If you don’t use perceptual info, some people call it contrast reduction.
Structured Face Hallucination Chih-Yuan Yang Sifei Liu Ming-Hsuan Yang Electrical Engineering and Computer Science 1.
Image Enhancement [DVT final project]
December 9, 2014Computer Vision Lecture 23: Motion Analysis 1 Now we will talk about… Motion Analysis.
Lec 22: Stereo CS4670 / 5670: Computer Vision Kavita Bala.
Computer Vision Lecture #10 Hossam Abdelmunim 1 & Aly A. Farag 2 1 Computer & Systems Engineering Department, Ain Shams University, Cairo, Egypt 2 Electerical.
The 18th Meeting on Image Recognition and Understanding 2015/7/29 Depth Image Enhancement Using Local Tangent Plane Approximations Kiyoshi MatsuoYoshimitsu.
Duy & Piotr. How to reconstruct a high quality image with the least amount of samples per pixel the least amount of resources And preserving the image.
Autonomous Robots Vision © Manfred Huber 2014.
Segmentation of Vehicles in Traffic Video Tun-Yu Chiang Wilson Lau.
Non-Ideal Iris Segmentation Using Graph Cuts
R I T Rochester Institute of Technology Geometric Scene Reconstruction Using 3-D Point Cloud Data Feng Li and Steve Lach Advanced Digital Image Processing.
Course14 Dynamic Vision. Biological vision can cope with changing world Moving and changing objects Change illumination Change View-point.
Yizhou Yu Texture-Mapping Real Scenes from Photographs Yizhou Yu Computer Science Division University of California at Berkeley Yizhou Yu Computer Science.
Demosaicking for Multispectral Filter Array (MSFA)
Jeong Kanghun CRV (Computer & Robot Vision) Lab..
CSCE 641 Computer Graphics: Image-based Rendering (cont.) Jinxiang Chai.
Speaker Min-Koo Kang March 26, 2013 Depth Enhancement Technique by Sensor Fusion: MRF-based approach.
Machine Vision Edge Detection Techniques ENT 273 Lecture 6 Hema C.R.
Performance Measurement of Image Processing Algorithms By Dr. Rajeev Srivastava ITBHU, Varanasi.
Projects Project 1a due this Friday Project 1b will go out on Friday to be done in pairs start looking for a partner now.
Digital Image Processing CSC331
Energy minimization Another global approach to improve quality of correspondences Assumption: disparities vary (mostly) smoothly Minimize energy function:
Fast edge-directed single-image super-resolution
Summary of “Efficient Deep Learning for Stereo Matching”
Semi-Global Matching with self-adjusting penalties
CS4670 / 5670: Computer Vision Kavita Bala Lec 27: Stereo.
Depth Map Upsampling by Self-Guided Residual Interpolation
A Gentle Introduction to Bilateral Filtering and its Applications
Energy Preserving Non-linear Filters
Range Imaging Through Triangulation
A Gentle Introduction to Bilateral Filtering and its Applications
HALO-FREE DESIGN FOR RETINEX BASED REAL-TIME VIDEO ENHANCEMENT SYSTEM
Presentation transcript:

Speaker Min-Koo Kang November 14, 2012 Depth Enhancement Technique by Sensor Fusion: Joint Bilateral Filter Approaches

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Outline 1. Introduction - Why depth data is important - How to acquire depth data - Depth upsampling: state-of-the-art approach 2. Background - Interpolation filters: Nearest Neighbor / Bilinear / Bicubic / Bilateral 3. Bilateral filter-based depth upsampling - Joint Bilateral Upsampling (JBU) filter / SIGGRAPH Pixel Weighted Average Strategy (PWAS) / ICIP Unified Multi-Lateral (UML) filter / AVSS Generalized depth enhancement framework / ECCV Concluding remarks

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion - Why depth data is important - How to acquire depth data - State-of-the-art approaches Introduction

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Why is depth data important?  Used in various fields  One of the most important techniques in computer vision  Important factors  speed, accuracy, resolution 3D reconstruction Virtual view generation In 3DTV Human computer interaction

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion How to acquire depth data?  Depth acquisition method comparison Laser scanning methodStereo vision sensorRange sensor  Can be overcome by depth map up-sampling Range sensor method has the most appropriate performance except low-resolution

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Problem definition  Disparity estimation by range sensor delivers small resolution of depth map  Rendering requires full resolution depth map  Main objectives / requirements: - Cost-effective (potential for real-time at consumer electronics platforms) - Align depth map edge with image edge - Remove inaccuracies (caused by heuristics in disparity estimation) - Temporal stability (esp. at edges and areas with detail) Upsampling Refinement

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Depth upsampling  Definition  Conversion of depth map with low resolution into one with high resolution  Approach  Most state-of-the-art methods are based on sensor fusion technique; i.e., use image sensor and range sensor together Depth map up-sampling by using bi-cubic interpolation Depth map up-sampling by using image and range sensor

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Background - Interpolation filters: Nearest Neighbor / Bilinear / Bicubic / Bilateral

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Single Image-based Interpolation  The conventional filterings

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion  The main types of artifacts are most easily seen at sharp edges, and include aliasing (jagged edges), blurring, and edge halos (see illustration below)  The main types of artifacts are most easily seen at sharp edges, and include aliasing (jagged edges), blurring, and edge halos (see illustration below) Upsampling examples 0% Sharpening 16.7% Sharpening 25% Sharpening Nearest NeighborBilinearBicubicInput

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Single Image-based Interpolation  Bilateral filtering: smoothing an image without blurring its edges

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Bilateral filtering applications InputGaussian smoothing Bilateral smoothing noisy image naïve denoising by Gaussian filter better denoising by bilateral filter

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Bilateral filter-based depth upsampling - Joint Bilateral Upsampling (JBU) filter / SIGGRAPH Pixel Weighted Average Strategy (PWAS) / ICIP Unified Multi-Lateral (UML) filter / AVSS Generalized depth enhancement framework / ECCV 2012

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Joint bilateral filtering  Multi-modal filtering  Range term defined by one modality  Filtering performed on an other modality  Propagates properties from one to an other modality  Edge preserving properties

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Joint bilateral upsampling (JBU)  First publication on bilateral filters for upsampling at SIGGRAPH 2007  J. Kopf, Univ. of Konstantz (Germany) provided reference sw.  [Kopf2007] solution:  High resolution image in range term  Low resolution input  high resolution output Kopf et al., “Joint Bilateral Upsampling”, SIGGRAPH 2007

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Joint bilateral upsampling (JBU)  Representative formulation:  N(P): targeting pixel P(i, j)’s neighborhood. f S (.): spatial weighting term, applied for pixel position P. f I (.): range weighting term, applied for pixel value I(q). f S (.), f I (.) are Gaussian functions with standard deviations, σ S and σ I, respectively. Kopf et al., “Joint Bilateral Upsampling”, SIGGRAPH 2007 Upsampled depth map Rendered 3D view

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Is JBU ideal enough?  Limitations of JBU:  It starts from the fundamental heuristic assumptions about the relationship between depth and intensity data  Sometimes depth has no corresponding edges in the 2-D image  Remaining problems:  Erroneous copying of 2-D texture into actually smooth geometries within the depth map  Unwanted artifact known as edge blurring High-resolution guidance image (red=non-visible depth discontinuities) Low-resolution depth map (red=zooming area) JBU enhanced depth map (zoomed)

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Pixel Weighted Average Strategy (PWAS)  Pixel Weighted Average Strategy for Depth Sensor Data Fusion  F. Garcia, proposed in ICIP 2010  [Garcia2010] solution:  Use of a *credibility map to cope with texture copy & edge blurring  Credibility map indicates unreliable regions in depth map  Representative formulation:  D: given depth map. Q: credibility map. Guiding intensity image. Garcia et al., “Pixel Weighted Average Strategy for Depth Sensor Data Fusion”, ICIP 2010 *credibility: 믿을 수 있음, 진실성; 신용, 신뢰성, 신빙성, 위신

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion High resolution image Low resolution depth JBU result PWAS result

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Again, is PWAS ideal enough?  Limitations of PWAS:  Degree of smoothing depends on gradient of low resolution depth map  Remaining problems:  Degree of smoothing depends on gradients of pixels in depth map  Erroneous depths around depth edge are not compensated well  Contradictive with spatial weight term (f S (.))  Texture copy issue still remains in homogeneous regions of depth map High-resolution guidance image (red=non-visible depth discontinuities) JBU enhanced depth map (zoomed) PWAS enhanced depth map (zoomed)

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Unified Multi-Lateral (UML) filter  In order to reduce texture copy issue, the same author proposed combined version of two PWAS  F. Garcia, proposed in AVSS 2011  [Garcia2011] solution:  Use of combined PWAS filters  The second filter has both spatial and range kernels acting onto D  Use of the credibility map Q as a blending function, i.e., β = Q  Representative formulation:  Depth pixels with high reliability are not influenced by the 2-D data avoiding texture copying Garcia et al., “A New Multilateral Filter for Real-Time Depth Enhancement”, AVSS 2011

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Depth map enhancement examples 2D guidance image JBUPWASUML 2D guidance image JBUPWASUML

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Again, is UML ideal enough?  Limitations of UML:  Features of the proposed filter strongly depends on the credibility map If reference pixel value in credibility map is low, The filter works as the normal PWAS filter by in order to reduce edge blurring artifact by weakening smoothing effect around depth edge. If reference pixel value in credibility map is high, Relatively high weigh is allocated to J 3, and the proposed filter works in direction of reducing texture copy artifact.  Remaining problems:  Is credibility map really credible?  It only considers depth gradient, but occlusion, shadowing, and homogeneous regions are really incredible in general depth data.  Edge blurring artifact still exists  when there’s no corresponding depth edge in the image due to similar object colors.

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Depth map enhancement examples Ground truth Downsampled (9x)Intensity image JBU PWASUML

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Generalized depth enhancement filter by sensor fusion  Generalize the previous UML filter not only for active sensors (RGB-D) but also more traditional stereo camera.  F. Garcia, proposed in ECCV 2012  [Garcia2012] solution:  Passive sensor: extension of credibility map for general depth data  Object boundary, occlusion, homogeneous regions are considered  Active sensor: adaptive blending function β(p) change to cope with edge blurring issue, and the second term (J 3 (p)) in UML is substituted by D(p)  Representative formulation:  Smoothing effect is reduced in credible depth regions  The same computational with PWAS complexity  New β(p) prevents edge blurring when image edges have similar color Garcia et al., “Generalized depth enhancement filter by sensor fusion”, ECCV 2012

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Generalized depth enhancement filter by sensor fusion  Formulation of a new credibility map (Q(p):  Boundary map Q b (p)  Q b (p) = Q(p) in J 2  Occlusion map Q o (p):  Homogeneous map Q h (p): the characteristics of correlation cost at each pixel is analyzed  Homogeneous region  flat correlation cost / repetitive pattern  multiple minima. cost  First minimum value at depth d1  C(p, d1) / second minimum at d2  C(p, d2) left/right consistency check Garcia et al., “Generalized depth enhancement filter by sensor fusion”, ECCV 2012

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Generalized depth enhancement filter by sensor fusion  Formulation of blending function β(p):  Q I is defined analogously to Q D but considering ∇ I  The function u(·) is a step function  If edge blurring condition is satisfied, β(p) = 1  i.e., Q D τ I  The constants τ I and τ D are empirically chosen thresholds  If not,  β(p) = Q D (p), and J 5 (p) works similarly to the conventional UML filter Garcia et al., “Generalized depth enhancement filter by sensor fusion”, ECCV 2012

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Experimental results – passive sensing Garcia et al., “Generalized depth enhancement filter by sensor fusion”, ECCV 2012

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion  RMS: Root Mean Square  PBMP : Percentage of Bad Matching Pixels  SSIM : Structural SIMilarity Experimental results – passive sensing Garcia et al., “Generalized depth enhancement filter by sensor fusion”, ECCV 2012

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Experimental results – active sensing ImageUIUI QDQD β Garcia et al., “Generalized depth enhancement filter by sensor fusion”, ECCV 2012

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Experimental results – active sensing Garcia et al., “Generalized depth enhancement filter by sensor fusion”, ECCV 2012

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Now then, do we have an optimal solution?  Limitations:  Initial depth 의 신뢰도가 낮을 경우에 image value 의 문제가 있으면 해당 position 의 depth 를 개선할 방법이 없다. 예를 들어, occlusion, homogeneous 영역에서 texture copying 문제가 여전히 발생 가능함.  UML filter 컨셉과 충동 ! Edge blurring 조건에서 depth edge 근처의 distortion 확산의 문제  Remaining problems:  Qb, Qo, Qh 의 역할이 완전히 독립적이지 못하기 때문에 over weighting 의 위험이 우려됨. 예를 들어, boundary 와 occlusion 영역이 겹치게 된다. 혹은 homogeneous 영역에서 잘못 추정된 depth 는 left/right consistency 결과 occlusion 영역으로 판단될 수도 있다.

/ Computer Vision Laboratory Seminar Depth enhancement technique by sensor fusion Conclusion  Joint bilateral upsampling approach  Propagates properties from one to an other modality  Credibility map decides system performance  Defining blending function can be another critical factor  Many empirical parameters make the practical automated usage of such fusion filter challenging  Another question is a clear rule on when a smoothing by filtering is to be avoided and when a simple binary decision is to be undertaken