An efficient method to build panoramic image mosaics

Similar documents
Multi-stable Perception. Necker Cube

A Binarization Algorithm specialized on Document Images and Photos

Computer Vision I. Xbox Kinnect: Rectification. The Fundamental matrix. Stereo III. CSE252A Lecture 16. Example: forward motion

Image warping and stitching May 5 th, 2015

Panorama Mosaic Optimization for Mobile Camera Systems

Image Alignment CSC 767

Structure from Motion

Content Based Image Retrieval Using 2-D Discrete Wavelet with Texture Feature with Different Classifiers

Feature Reduction and Selection

Feature-based image registration using the shape context

Computer Animation and Visualisation. Lecture 4. Rigging / Skinning

Image Fusion With a Dental Panoramic X-ray Image and Face Image Acquired With a KINECT

Outline. Seamless Image Stitching in the Gradient Domain. Related Approaches. Image Stitching. Introduction Related Work

Ecient Computation of the Most Probable Motion from Fuzzy. Moshe Ben-Ezra Shmuel Peleg Michael Werman. The Hebrew University of Jerusalem

A Robust Method for Estimating the Fundamental Matrix

An Image Fusion Approach Based on Segmentation Region

Corner-Based Image Alignment using Pyramid Structure with Gradient Vector Similarity

Learning the Kernel Parameters in Kernel Minimum Distance Classifier

Calibrating a single camera. Odilon Redon, Cyclops, 1914

Range Data Registration Using Photometric Features

Improvement of Spatial Resolution Using BlockMatching Based Motion Estimation and Frame. Integration

A Fast Visual Tracking Algorithm Based on Circle Pixels Matching

What are the camera parameters? Where are the light sources? What is the mapping from radiance to pixel color? Want to solve for 3D geometry

MOTION PANORAMA CONSTRUCTION FROM STREAMING VIDEO FOR POWER- CONSTRAINED MOBILE MULTIMEDIA ENVIRONMENTS XUNYU PAN

IMAGE STITCHING WITH PERSPECTIVE-PRESERVING WARPING

Real-time Motion Capture System Using One Video Camera Based on Color and Edge Distribution

Fitting & Matching. Lecture 4 Prof. Bregler. Slides from: S. Lazebnik, S. Seitz, M. Pollefeys, A. Effros.

PERFORMANCE EVALUATION FOR SCENE MATCHING ALGORITHMS BY SVM

Semarang, Indonesia. Sepuluh Nopember Institute of Technology, Surabaya, Indonesia

Inverse-Polar Ray Projection for Recovering Projective Transformations

Homography-Based 3D Scene Analysis of Video Sequences *

Scan Conversion & Shading

MULTISPECTRAL IMAGES CLASSIFICATION BASED ON KLT AND ATR AUTOMATIC TARGET RECOGNITION

Scan Conversion & Shading

Local Quaternary Patterns and Feature Local Quaternary Patterns

Fitting and Alignment

Skew Angle Estimation and Correction of Hand Written, Textual and Large areas of Non-Textual Document Images: A Novel Approach

FEATURE EXTRACTION. Dr. K.Vijayarekha. Associate Dean School of Electrical and Electronics Engineering SASTRA University, Thanjavur

THE RENDERING OF BUILDING TEXTURE FROM LAND-BASED VIDEO

Detecting Maximum Inscribed Rectangle Based On Election Campaign Algorithm Qing-Hua XIE1,a,*, Xiang-Wei ZHANG1,b, Wen-Ge LV1,c and Si-Yuan CHENG1,d

Invariant Shape Object Recognition Using B-Spline, Cardinal Spline, and Genetic Algorithm

A Fast Content-Based Multimedia Retrieval Technique Using Compressed Data

Simulation: Solving Dynamic Models ABE 5646 Week 11 Chapter 2, Spring 2010

Shape Representation Robust to the Sketching Order Using Distance Map and Direction Histogram

Acoustic Camera Image Mosaicing and Super-resolution

Data Modelling and. Multimedia. Databases M. Multimedia. Information Retrieval Part II. Outline

Angle-Independent 3D Reconstruction. Ji Zhang Mireille Boutin Daniel Aliaga

Development of an Active Shape Model. Using the Discrete Cosine Transform

New dynamic zoom calibration technique for a stereo-vision based multi-view 3D modeling system

Multi-view 3D Position Estimation of Sports Players

Computer Generated Integral Imaging (II) System Using Depth-Camera

Reducing Frame Rate for Object Tracking

MOTION BLUR ESTIMATION AT CORNERS

Online codebook modeling based background subtraction with a moving camera

ScienceDirect. The Influence of Subpixel Corner Detection to Determine the Camera Displacement

Collaboratively Regularized Nearest Points for Set Based Recognition

Some Tutorial about the Project. Computer Graphics

Line-based Camera Movement Estimation by Using Parallel Lines in Omnidirectional Video

A Range Image Refinement Technique for Multi-view 3D Model Reconstruction

Problem Definitions and Evaluation Criteria for Computational Expensive Optimization

Model-Based Bundle Adjustment to Face Modeling

Subspace clustering. Clustering. Fundamental to all clustering techniques is the choice of distance measure between data points;

Smoothing Spline ANOVA for variable screening

3D vector computer graphics

Mobile Robot Localization and Mapping by Scan Matching using Laser Reflection Intensity of the SOKUIKI Sensor

Integrated Expression-Invariant Face Recognition with Constrained Optical Flow

Geometric Transformations and Multiple Views

Parallelism for Nested Loops with Non-uniform and Flow Dependences

Object Tracking Based on PISC Image and Template Matching

Direct Methods for Visual Scene Reconstruction

SLAM Summer School 2006 Practical 2: SLAM using Monocular Vision

Suppression for Luminance Difference of Stereo Image-Pair Based on Improved Histogram Equalization

High resolution 3D Tau-p transform by matching pursuit Weiping Cao* and Warren S. Ross, Shearwater GeoServices

An Improved Image Segmentation Algorithm Based on the Otsu Method

Scale Selective Extended Local Binary Pattern For Texture Classification

Machine Learning: Algorithms and Applications

Finite Element Analysis of Rubber Sealing Ring Resilience Behavior Qu Jia 1,a, Chen Geng 1,b and Yang Yuwei 2,c

Fingerprint matching based on weighting method and SVM

Straight Line Detection Based on Particle Swarm Optimization

The Research of Ellipse Parameter Fitting Algorithm of Ultrasonic Imaging Logging in the Casing Hole

Face Recognition University at Buffalo CSE666 Lecture Slides Resources:

Using Fuzzy Logic to Enhance the Large Size Remote Sensing Images

Grading Image Retrieval Based on DCT and DWT Compressed Domains Using Low-Level Features

IMAGE MATCHING WITH SIFT FEATURES A PROBABILISTIC APPROACH

An Entropy-Based Approach to Integrated Information Needs Assessment

Environmental Mapping by Trinocular Vision for Self-Localization Using Monocular Vision

Real-time Feature-based Video Mosaicing at 500 fps

EYE CENTER LOCALIZATION ON A FACIAL IMAGE BASED ON MULTI-BLOCK LOCAL BINARY PATTERNS

Classifier Selection Based on Data Complexity Measures *

User Authentication Based On Behavioral Mouse Dynamics Biometrics

UAV global pose estimation by matching forward-looking aerial images with satellite images

Maximum Variance Combined with Adaptive Genetic Algorithm for Infrared Image Segmentation

Computer Graphics. Jeng-Sheng Yeh 葉正聖 Ming Chuan University (modified from Bing-Yu Chen s slides)

An Automatic Eye Detection Method for Gray Intensity Facial Images

A Clustering Algorithm for Key Frame Extraction Based on Density Peak

Geometric Primitive Refinement for Structured Light Cameras

Image Representation & Visualization Basic Imaging Algorithms Shape Representation and Analysis. outline

A Novel Adaptive Descriptor Algorithm for Ternary Pattern Textures

PROJECTIVE RECONSTRUCTION OF BUILDING SHAPE FROM SILHOUETTE IMAGES ACQUIRED FROM UNCALIBRATED CAMERAS

Fuzzy Logic Based RS Image Classification Using Maximum Likelihood and Mahalanobis Distance Classifiers

Transcription:

An effcent method to buld panoramc mage mosacs Pattern Recognton Letters vol. 4 003 Dae-Hyun Km Yong-In Yoon Jong-Soo Cho School of Electrcal Engneerng and Computer Scence Kyungpook Natonal Unv.

Abstract Proposed method Computng the projectve transformaton n overlapped areas of the two gven mages Usng four seed ponts /

Introducton A number of methods to buld mage mosacs Recordng an mage onto a long flmstrp Usng the panorama camera Usng lens wth a very large feld of vew Fsh-eye lens and parabolc mrror Image mosackng algorthm Takng regular pctures n order to cover the whole vewng space Algnng these mages and puttng them together Conventonal algorthm Usng geometrcal feature ponts and optmzaton to compute the projectve transformaton Consumng very tme because of the teratve computaton 3 /

Proposed method Usng four pars of seed ponts to compute the projectve transform n the overlapped areas of two mages Seed pont s the hghest textured pxel n the overlapped area of the reference mage Detecton of the correspondng pont Usng block matchng algorthm(bma) n the overlapped area of the target mage For more robust detecton of the correspondng pont Hstogram equalzaton of two overlapped mages before selectng seed ponts Addng the weght functon to BMA n order to mnmze mage dstorton due to camera rotaton 4 /

Flow chart Fg. 1. The proposed algorthm. 5 /

Computaton of the projectve transform Extractng overlapped areas Usng phase correlaton Algnng two mages that are shfted relatvely to each other Takng D fourer transform of each mage Computng the phase dfference at each frequency Takng the nverse fourer transform Estmatng overlapped areas of two gven mages Selectng seed ponts Dvdng overlapped area of the reference mage nto four subareas Central pxel of the block that kept the maxmum varance n each subarea 6 /

q σ = k j arg max = G MAX g g k [ ] σ h g k 0 3 M k j (1) σ k where varance M k mean value of the kth block n th subarea h g hstogram of the graylevel g maxmum graylevel G MAX Detectng correspondng ponts Hstogram equalzaton Comparng equally wthout the bas generatng due to the dfference of the perceved contrast and brghtness 7 /

8 / / More robust detecton of correspondences Usng weghted BMA Consderaton of the dstorton generatng due to camera rotaton j B j y x y x w k d j y d x I k j y x I d d E ) ( 1) ( ) ( ) ( + + + + + + + = () (3) 7 7 / j d j D d w j j j + = = where graylevel of (xy) n the kth mage dsplacement of the block dstance from the center of block maxmum dstance ) ( k y x I x d y d j d D

Buldng the panoramc mage mosac Transformaton of cylndrcal coordnates Estmatng the focal length from the projectve transformaton By Szelsk and Shum (1997) Warpng each perspectve mage nto cylndrcal coordnates World coordnate P( X Y Z) D cylndrcal screen coordnate ( θ v) θ = tan 1 where θ pannng angle v scanlne ( X / Z) v = Y / X + Z (4) 9 /

Because only the screen coordnate p( x y) Eq. (4) transform of screen coordnates Camera equatons x = f X / Y y = f Y / Z θ = tan 1 ( x / f ) v = y / x + f (5) Blendng warped mages Estmatng the dsplacement between two cylndrcally warped mages Usng phase correlaton Constructng panoramc mage Usng blnear weghtng functon 10 /

Expermentaton Expermental condtons and test mages 15x15 block and 3x3 search range (a) (b) (c) Fg.. Expermental mage pars. (d) 11 /

Estmatng PSNR n order to evaluate the performance of the proposed algorthm Accuracy of projectve transform Comparson by hstogram equalzaton Table 1. Means and standard devaton before and after hstogram equalzaton. 1 /

(a) (b) Fg. 3. Two overlapped mages and ther hstograms before (a) and after (b) hstogram equalzaton. 13 /

PSNR of the reconstructed mage mosacs before and after hstogram equalzaton Fg. 4. Performance comparson before and after hstogram equalzaton. 14 /

Weghted block matchng algorthm Fg. 5. Performance comparson between typcal BMA and weghted BMA. 15 /

Ghost mage phenomenon (a) Fg. 6. Mosackng mages of Fg. (b): (a) typcal BMA and (b) weghted BMA. (b) 16 /

Comparson by the mage mosackng algorthms Imad s algorthm Usng the geometrcal feature ponts Szelsk s algorthm Usng the Levenberg-Marquardt teratve optmzaton Proposed algorthm 17 /

(a) (b) (c) (d) Fg. 7. Mosackng mages: (a) mage mosac by human anmator (b) mage mosac by Szelsk s algorthm (c) mage mosac by the proposed algorthm and (d) PSNR of each algorthm. 18 /

Further analyss of the proposed algorthm Mnmzng the sum of the squared ntensty error E = [ ] ' ' I'( x y ) I( x y ) = where I( x y ) reference mage ' ' ' I ( x y ) warped target mage e ' (5) Table. Features for computng the projectve transform. Table 3. Quanttatve comparson between Szelsk s algorthm and the proposed algorthm. 19 /

(a) (b) Fg. 8. Reconstructed mage mosacs 0 /

(a) (b) Fg. 9. Reconstructed cylndrcal mage mosacs. 1 /

Concluson Proposed method An effcent method to buld a panoramc mage mosac wth four seed ponts Selecton of the hghest textured pxel n each subarea Detecton of correspondence Usng the weghted BMA n order to mnmze mage dstorton Future work Vew nterpolaton Generatng the novel scene n accordance wth the movement of the vewer s gazng drecton /