1 Faculty of Information Technology Enhanced Generic Fourier Descriptor for Object-Based Image Retrieval Dengsheng Zhang, Guojun Lu Gippsland School of.

Slides:



Advertisements
Similar presentations
Image Registration  Mapping of Evolution. Registration Goals Assume the correspondences are known Find such f() and g() such that the images are best.
Advertisements

Context-based object-class recognition and retrieval by generalized correlograms by J. Amores, N. Sebe and P. Radeva Discussion led by Qi An Duke University.
Wavelets Fast Multiresolution Image Querying Jacobs et.al. SIGGRAPH95.
November 12, 2013Computer Vision Lecture 12: Texture 1Signature Another popular method of representing shape is called the signature. In order to compute.
Extended Gaussian Images
A NOVEL LOCAL FEATURE DESCRIPTOR FOR IMAGE MATCHING Heng Yang, Qing Wang ICME 2008.
3D Shape Histograms for Similarity Search and Classification in Spatial Databases. Mihael Ankerst,Gabi Kastenmuller, Hans-Peter-Kriegel,Thomas Seidl Univ.
Chapter 8 Content-Based Image Retrieval. Query By Keyword: Some textual attributes (keywords) should be maintained for each image. The image can be indexed.
IDEAL 2005, 6-8 July, Brisbane Multiresolution Analysis of Connectivity Atul Sajjanhar, Deakin University, Australia Guojun Lu, Monash University, Australia.
PHP-based Image Recognition and Retrieval of Late 18th Century Artwork Ben Goodwin Handouts are available for students writing summaries for class assignments.
Image Indexing and Retrieval using Moment Invariants Imran Ahmad School of Computer Science University of Windsor – Canada.
Xianfeng Gu, Yaling Wang, Tony Chan, Paul Thompson, Shing-Tung Yau
Vicenç Parisi Baradad, Joan Cabestany, Jaume Piera
November 4, 2014Computer Vision Lecture 15: Shape Representation II 1Signature Another popular method of representing shape is called the signature. In.
ICIP 2000, Vancouver, Canada IVML, ECE, NTUA Face Detection: Is it only for Face Recognition?  A few years earlier  Face Detection Face Recognition 
Young Deok Chun, Nam Chul Kim, Member, IEEE, and Ick Hoon Jang, Member, IEEE IEEE TRANSACTIONS ON MULTIMEDIA,OCTOBER 2008.
A Study of Approaches for Object Recognition
1 Visual Information Extraction in Content-based Image Retrieval System Presented by: Mian Huang Weichuan Dong Apr 29, 2004.
Detecting Image Region Duplication Using SIFT Features March 16, ICASSP 2010 Dallas, TX Xunyu Pan and Siwei Lyu Computer Science Department University.
Digital Days 29/6/2001 ISTORAMA: A Content-Based Image Search Engine and Hierarchical Triangulation of 3D Surfaces. Dr. Ioannis Kompatsiaris Centre for.
Automatic Image Alignment (feature-based) : Computational Photography Alexei Efros, CMU, Fall 2005 with a lot of slides stolen from Steve Seitz and.
Visual Querying By Color Perceptive Regions Alberto del Bimbo, M. Mugnaini, P. Pala, and F. Turco University of Florence, Italy Pattern Recognition, 1998.
Stepan Obdrzalek Jirı Matas
Automatic Image Alignment (feature-based) : Computational Photography Alexei Efros, CMU, Fall 2006 with a lot of slides stolen from Steve Seitz and.
Ashish Uthama EOS 513 Term Paper Presentation Ashish Uthama Biomedical Signal and Image Computing Lab Department of Electrical.
Content-based Image Retrieval (CBIR)
SIEVE—Search Images Effectively through Visual Elimination Ying Liu, Dengsheng Zhang and Guojun Lu Gippsland School of Info Tech,
Introduction --Classification Shape ContourRegion Structural Syntactic Graph Tree Model-driven Data-driven Perimeter Compactness Eccentricity.
AdvisorStudent Dr. Jia Li Shaojun Liu Dept. of Computer Science and Engineering, Oakland University 3D Shape Classification Using Conformal Mapping In.
Wavelet-Based Multiresolution Matching for Content-Based Image Retrieval Presented by Tienwei Tsai Department of Computer Science and Engineering Tatung.
1 Faculty of Information Technology Generic Fourier Descriptor for Shape-based Image Retrieval Dengsheng Zhang, Guojun Lu Gippsland School of Comp. & Info.
Recognition and Matching based on local invariant features Cordelia Schmid INRIA, Grenoble David Lowe Univ. of British Columbia.
Alignment and Matching
Digital Image Processing, 2nd ed. © 2002 R. C. Gonzalez & R. E. Woods Chapter 11 Representation & Description Chapter 11 Representation.
Digital Image Processing Lecture 20: Representation & Description
Shape Matching with Occlusion in Image Databases Aristeidis Diplaros Euripides G.M. Petrakis Evangelos Milios Technical University of Crete.
Shape Based Image Retrieval Using Fourier Descriptors Dengsheng Zhang and Guojun Lu Gippsland School of Computing and Information Technology Monash University.
A Statistical Approach to Speed Up Ranking/Re-Ranking Hong-Ming Chen Advisor: Professor Shih-Fu Chang.
A 3D Model Alignment and Retrieval System Ding-Yun Chen and Ming Ouhyoung.
Shape Analysis and Retrieval Structural Shape Descriptors Notes courtesy of Funk et al., SIGGRAPH 2004.
COLOR HISTOGRAM AND DISCRETE COSINE TRANSFORM FOR COLOR IMAGE RETRIEVAL Presented by 2006/8.
Picture Comparison; now with shapes!
Generalized Hough Transform
Digital Image Processing, 2nd ed. © 2002 R. C. Gonzalez & R. E. Woods Representation & Description.
Content-Based Image Retrieval Using Fuzzy Cognition Concepts Presented by Tienwei Tsai Department of Computer Science and Engineering Tatung University.
Lecture 7: Features Part 2 CS4670/5670: Computer Vision Noah Snavely.
Dengsheng Zhang and Melissa Chen Yi Lim
Fourier Descriptors For Shape Recognition Applied to Tree Leaf Identification By Tyler Karrels.
2005/12/021 Content-Based Image Retrieval Using Grey Relational Analysis Dept. of Computer Engineering Tatung University Presenter: Tienwei Tsai ( 蔡殿偉.
Event retrieval in large video collections with circulant temporal encoding CVPR 2013 Oral.
2005/12/021 Fast Image Retrieval Using Low Frequency DCT Coefficients Dept. of Computer Engineering Tatung University Presenter: Yo-Ping Huang ( 黃有評 )
Content-Based Image Retrieval Using Block Discrete Cosine Transform Presented by Te-Wei Chiang Department of Information Networking Technology Chihlee.
CS654: Digital Image Analysis Lecture 36: Feature Extraction and Analysis.
CS848 Similarity Search in Multimedia Databases Dr. Gisli Hjaltason Content-based Retrieval Using Local Descriptors: Problems and Issues from Databases.
Content-Based Image Retrieval QBIC Homepage The State Hermitage Museum db2www/qbicSearch.mac/qbic?selLang=English.
A Distributed Multimedia Data Management over the Grid Kasturi Chatterjee Advisors for this Project: Dr. Shu-Ching Chen & Dr. Masoud Sadjadi Distributed.
Content Based Color Image Retrieval vi Wavelet Transformations Information Retrieval Class Presentation May 2, 2012 Author: Mrs. Y.M. Latha Presenter:
An Image Retrieval Approach Based on Dominant Wavelet Features Presented by Te-Wei Chiang 2006/4/1.
Partial Shape Matching. Outline: Motivation Sum of Squared Distances.
Image features and properties. Image content representation The simplest representation of an image pattern is to list image pixels, one after the other.
Sheng-Fang Huang Chapter 11 part I.  After the image is segmented into regions, how to represent and describe these regions? ◦ In terms of its external.
Image Representation and Description – Representation Schemes
IT472: Digital Image Processing
Digital Image Processing Lecture 20: Representation & Description
Content-Based Image Retrieval Readings: Chapter 8:
Slope and Curvature Density Functions
Improving Retrieval Performance of Zernike Moment Descriptor on Affined Shapes Dengsheng Zhang, Guojun Lu Gippsland School of Comp. & Info Tech Monash.
Fourier Transform of Boundaries
Recognition and Matching based on local invariant features
Presentation transcript:

1 Faculty of Information Technology Enhanced Generic Fourier Descriptor for Object-Based Image Retrieval Dengsheng Zhang, Guojun Lu Gippsland School of Comp. & Info Tech Monash University Churchill, VIC 3842 Australia

2 Faculty of Information Technology Outline Motivations Generic Fourier Descriptor (GFD) Problem Enhanced Generic Fourier Descriptor (EGFD) Experimental Results Conclusions Motivations Generic Fourier Descriptor (GFD) Problem Enhanced Generic Fourier Descriptor (EGFD) Experimental Results Conclusions

3 Faculty of Information Technology Motivations Content-based Image Retrieval –Shape is an important image feature along with color and texture Effective and Efficient Shape Descriptor –good retrieval accuracy, compact features, general application, low computation complexity, robust retrieval performance and hierarchical coarse to fine representation Affined Shape Retrieval –Affined shapes are common in nature due to objects being viewed from different angles and objects being stretched, skewed. Content-based Image Retrieval –Shape is an important image feature along with color and texture Effective and Efficient Shape Descriptor –good retrieval accuracy, compact features, general application, low computation complexity, robust retrieval performance and hierarchical coarse to fine representation Affined Shape Retrieval –Affined shapes are common in nature due to objects being viewed from different angles and objects being stretched, skewed.

4 Faculty of Information Technology Affine Distorted Shapes Are Common

5 Faculty of Information Technology Generic Fourier Descriptor Polar Transform –For an input image f(x, y), it is first transformed into polar image f(r,  ): Polar Transform –For an input image f(x, y), it is first transformed into polar image f(r,  ):

6 Faculty of Information Technology Generic Fourier Descriptor-II Polar Raster Sampling Polar Grid Polar image Polar raster sampled image in Cartesian space

7 Faculty of Information Technology Generic Fourier Descriptor-III 2-D Fourier transform on polar raster sampled image f(r,  ): where 0  r<R and  i = i(2  /T) (0  i<T); 0  <R, 0  <T. R and T are the radial frequency resolution and angular frequency resolution respectively. The normalized Fourier coefficients are the GFD. 2-D Fourier transform on polar raster sampled image f(r,  ): where 0  r<R and  i = i(2  /T) (0  i<T); 0  <R, 0  <T. R and T are the radial frequency resolution and angular frequency resolution respectively. The normalized Fourier coefficients are the GFD.

8 Faculty of Information Technology Problem Generally, GFD has good performance on generic shapes. Its overall retrieval precision after full recall is 98.6% for rotated shapes, 90.5% for scaled shapes, 74.1% for perspective transformed shapes and 80.5% for generally distorted shapes. Compared with rotation and scaling invariance test, the retrieval performance on perspective transform and generally distorted shapes are significantly lower. The problem is caused by the polar raster sampling method. Generally, GFD has good performance on generic shapes. Its overall retrieval precision after full recall is 98.6% for rotated shapes, 90.5% for scaled shapes, 74.1% for perspective transformed shapes and 80.5% for generally distorted shapes. Compared with rotation and scaling invariance test, the retrieval performance on perspective transform and generally distorted shapes are significantly lower. The problem is caused by the polar raster sampling method.

9 Faculty of Information Technology Under-sampling Problem Only half the sampled positions contain shape information

10 Faculty of Information Technology Enhanced GFD Normalization –Find major axis –Rotation normalization so that major axis of the shape is horizontal –Scaling normalization so that the ecentricity of the shape is 1. Normalization –Find major axis –Rotation normalization so that major axis of the shape is horizontal –Scaling normalization so that the ecentricity of the shape is 1.

11 Faculty of Information Technology Enhanced GFD-II Optimized Major Axis Algorithm (MAA) –Find the boundary point pairs in a number of directions (e.g. 360). –Find the two points p 1, p 2 with the furthest distance in the found boundary points, then p 1 p 2 is the major axis. The computation of MAA is O(N) rather than O(N 2 ). Optimized Major Axis Algorithm (MAA) –Find the boundary point pairs in a number of directions (e.g. 360). –Find the two points p 1, p 2 with the furthest distance in the found boundary points, then p 1 p 2 is the major axis. The computation of MAA is O(N) rather than O(N 2 ).

12 Faculty of Information Technology Enhanced GFD-III Normalization Result:

13 Faculty of Information Technology Enhanced GFD-IV Applying GFD transform on the rotation and scaling normalized image. The normalized transform coefficients are the enhanced GFD (EGFD). Applying GFD transform on the rotation and scaling normalized image. The normalized transform coefficients are the enhanced GFD (EGFD).

14 Faculty of Information Technology Experiment Dataset –Two datasets from MPEG-7 region shape database CE-2 are used. (CE-2 has been organized by MPEG-7 into six datasets to test a shape descriptor’s behaviors under different distortions) –Set A4 consists of 3101 from the whole database, it is for test of robustness to perspective transform. 330 shapes in Set A4 have been organized into 30 groups (11 similar shapes in each group) which are used as queries. –The whole database consists of 3621 shapes, 651 shapes have been organized into 31 groups (21 similar shapes in each group) which are used as queries. Indexing and automatic retrieval Dataset –Two datasets from MPEG-7 region shape database CE-2 are used. (CE-2 has been organized by MPEG-7 into six datasets to test a shape descriptor’s behaviors under different distortions) –Set A4 consists of 3101 from the whole database, it is for test of robustness to perspective transform. 330 shapes in Set A4 have been organized into 30 groups (11 similar shapes in each group) which are used as queries. –The whole database consists of 3621 shapes, 651 shapes have been organized into 31 groups (21 similar shapes in each group) which are used as queries. Indexing and automatic retrieval

15 Faculty of Information Technology Performance Measurement Recall vs Precision

16 Faculty of Information Technology Results Recall-Precision of EGFD on perspective shapes. –Compared with GFD, the improvement on Set A4 is 15.4%, the overall precision is increased from 74.1% to 89.5%. Recall-Precision of EGFD on perspective shapes. –Compared with GFD, the improvement on Set A4 is 15.4%, the overall precision is increased from 74.1% to 89.5%.

17 Faculty of Information Technology Results Recall-Precision of EGFD on Generally Distorted Shapes –Compared with GFD, the improvement on CE-2 is 12%, the overall precision is increased from 80.5% to 92.5%. Recall-Precision of EGFD on Generally Distorted Shapes –Compared with GFD, the improvement on CE-2 is 12%, the overall precision is increased from 80.5% to 92.5%.

18 Faculty of Information Technology Results EGFD GFD ZMD

19 Faculty of Information Technology Results EGFD GFD ZMD

20 Faculty of Information Technology Application of EGFD The application of the enhancement process is database/ application dependent. If the database has abundant perspective shapes, this technique can be very effective in retrieving similar shapes. If the database does not have perspective shapes, or the user wants finer distinction between similar shapes, the enhanced process may not be desirable. For example, if the user wants to distinguish between rectangles and squares, or to distinguish between ellipses and circles, the enhanced GFD can fail, because it normalizes all the shapes into same eccentricity (=1). In general applications, the enhancement is a useful option to the retrieval system rather than the replacement of GFD. The application of the enhancement process is database/ application dependent. If the database has abundant perspective shapes, this technique can be very effective in retrieving similar shapes. If the database does not have perspective shapes, or the user wants finer distinction between similar shapes, the enhanced process may not be desirable. For example, if the user wants to distinguish between rectangles and squares, or to distinguish between ellipses and circles, the enhanced GFD can fail, because it normalizes all the shapes into same eccentricity (=1). In general applications, the enhancement is a useful option to the retrieval system rather than the replacement of GFD.

21 Faculty of Information Technology Conclusions The proposed EGFD improves GFD significantly. It improves GFD’s relatively lower retrieval performance on severely skewed and stretched shapes. It also improves GFD’s robustness to general shape distortions. A shape normalization method is presented. The shape normalization method can be exploited for general shape representation purposes. An optimized major axis algorithm (MAA) is proposed. MA is a common normalization mechanism in shape modeling and representation. Common MAA is only for finding MA of contour shape. The proposed optimized MAA can be used for finding MA of generic shapes. The proposed EGFD improves GFD significantly. It improves GFD’s relatively lower retrieval performance on severely skewed and stretched shapes. It also improves GFD’s robustness to general shape distortions. A shape normalization method is presented. The shape normalization method can be exploited for general shape representation purposes. An optimized major axis algorithm (MAA) is proposed. MA is a common normalization mechanism in shape modeling and representation. Common MAA is only for finding MA of contour shape. The proposed optimized MAA can be used for finding MA of generic shapes.