Using eigencolor normalization for illumination-invariant color object recognition Speaker: 鄭雅勻 Date:2010/12/30 Zhenyong Lin, Junxian Wang, Kai-Kuang Ma.

Slides:



Advertisements
Similar presentations
3D Model Matching with Viewpoint-Invariant Patches(VIP) Reporter :鄒嘉恆 Date : 10/06/2009.
Advertisements

Order Structure, Correspondence, and Shape Based Categories Presented by Piotr Dollar October 24, 2002 Stefan Carlsson.
QR Code Recognition Based On Image Processing
Road-Sign Detection and Recognition Based on Support Vector Machines Saturnino, Sergio et al. Yunjia Man ECG 782 Dr. Brendan.
Presented by Xinyu Chang
Evaluating Color Descriptors for Object and Scene Recognition Koen E.A. van de Sande, Student Member, IEEE, Theo Gevers, Member, IEEE, and Cees G.M. Snoek,
電腦視覺 Computer and Robot Vision I
Chapter 3 Image Enhancement in the Spatial Domain.
Hongliang Li, Senior Member, IEEE, Linfeng Xu, Member, IEEE, and Guanghui Liu Face Hallucination via Similarity Constraints.
Extended Gaussian Images
Image Processing IB Paper 8 – Part A Ognjen Arandjelović Ognjen Arandjelović
Low Complexity Keypoint Recognition and Pose Estimation Vincent Lepetit.
Automatic Feature Extraction for Multi-view 3D Face Recognition
Mapping: Scaling Rotation Translation Warp
A Versatile Depalletizer of Boxes Based on Range Imagery Dimitrios Katsoulas*, Lothar Bergen*, Lambis Tassakos** *University of Freiburg **Inos Automation-software.
ECCV 2002 Removing Shadows From Images G. D. Finlayson 1, S.D. Hordley 1 & M.S. Drew 2 1 School of Information Systems, University of East Anglia, UK 2.
ICCV 2003 Colour Workshop 1 Recovery of Chromaticity Image Free from Shadows via Illumination Invariance Mark S. Drew 1, Graham D. Finlayson 2, & Steven.
Object Recognition with Invariant Features n Definition: Identify objects or scenes and determine their pose and model parameters n Applications l Industrial.
ICIP 2000, Vancouver, Canada IVML, ECE, NTUA Face Detection: Is it only for Face Recognition?  A few years earlier  Face Detection Face Recognition 
Modeling Pixel Process with Scale Invariant Local Patterns for Background Subtraction in Complex Scenes (CVPR’10) Shengcai Liao, Guoying Zhao, Vili Kellokumpu,
Chapter 5 Orthogonality
A Study of Approaches for Object Recognition
Uncalibrated Geometry & Stratification Sastry and Yang
COMP322/S2000/L221 Relationship between part, camera, and robot (cont’d) the inverse perspective transformation which is dependent on the focal length.
CS292 Computational Vision and Language Visual Features - Colour and Texture.
1 Invariant Local Feature for Object Recognition Presented by Wyman 2/05/2006.
Smart Traveller with Visual Translator for OCR and Face Recognition LYU0203 FYP.
Introduction of the intrinsic image. Intrinsic Images The method of Finlayson & Hordley ( 2001 ) Two assumptions 1. the camera ’ s sensors are sufficiently.
Statistical Color Models (SCM) Kyungnam Kim. Contents Introduction Trivariate Gaussian model Chromaticity models –Fixed planar chromaticity models –Zhu.
AdvisorStudent Dr. Jia Li Shaojun Liu Dept. of Computer Science and Engineering, Oakland University 3D Shape Classification Using Conformal Mapping In.
HMM-BASED PSEUDO-CLEAN SPEECH SYNTHESIS FOR SPLICE ALGORITHM Jun Du, Yu Hu, Li-Rong Dai, Ren-Hua Wang Wen-Yi Chu Department of Computer Science & Information.
Tricolor Attenuation Model for Shadow Detection. INTRODUCTION Shadows may cause some undesirable problems in many computer vision and image analysis tasks,
Matching 3D Shapes Using 2D Conformal Representations Xianfeng Gu 1, Baba Vemuri 2 Computer and Information Science and Engineering, Gainesville, FL ,
Wavelet-Based Multiresolution Matching for Content-Based Image Retrieval Presented by Tienwei Tsai Department of Computer Science and Engineering Tatung.
Recognition and Matching based on local invariant features Cordelia Schmid INRIA, Grenoble David Lowe Univ. of British Columbia.
BACKGROUND LEARNING AND LETTER DETECTION USING TEXTURE WITH PRINCIPAL COMPONENT ANALYSIS (PCA) CIS 601 PROJECT SUMIT BASU FALL 2004.
The Brightness Constraint
Course 12 Calibration. 1.Introduction In theoretic discussions, we have assumed: Camera is located at the origin of coordinate system of scene.
Simple Image Processing Speaker : Lin Hsiu-Ting Date : 2005 / 04 / 27.
A 3D Model Alignment and Retrieval System Ding-Yun Chen and Ming Ouhyoung.
视觉的三维运动理解 刘允才 上海交通大学 2002 年 11 月 16 日 Understanding 3D Motion from Images Yuncai Liu Shanghai Jiao Tong University November 16, 2002.
COLOR HISTOGRAM AND DISCRETE COSINE TRANSFORM FOR COLOR IMAGE RETRIEVAL Presented by 2006/8.
Correspondence-Free Determination of the Affine Fundamental Matrix (Tue) Young Ki Baik, Computer Vision Lab.
Phase Congruency Detects Corners and Edges Peter Kovesi School of Computer Science & Software Engineering The University of Western Australia.
In Defense of Nearest-Neighbor Based Image Classification Oren Boiman The Weizmann Institute of Science Rehovot, ISRAEL Eli Shechtman Adobe Systems Inc.
Dengsheng Zhang and Melissa Chen Yi Lim
2005/12/021 Content-Based Image Retrieval Using Grey Relational Analysis Dept. of Computer Engineering Tatung University Presenter: Tienwei Tsai ( 蔡殿偉.
2005/12/021 Fast Image Retrieval Using Low Frequency DCT Coefficients Dept. of Computer Engineering Tatung University Presenter: Yo-Ping Huang ( 黃有評 )
Geometric Hashing: A General and Efficient Model-Based Recognition Scheme Yehezkel Lamdan and Haim J. Wolfson ICCV 1988 Presented by Budi Purnomo Nov 23rd.
A Flexible New Technique for Camera Calibration Zhengyou Zhang Sung Huh CSPS 643 Individual Presentation 1 February 25,
A NOVEL METHOD FOR COLOR FACE RECOGNITION USING KNN CLASSIFIER
Content-Based Image Retrieval Using Block Discrete Cosine Transform Presented by Te-Wei Chiang Department of Information Networking Technology Chihlee.
Last update Heejune Ahn, SeoulTech
1 Chapter 2: Geometric Camera Models Objective: Formulate the geometrical relationships between image and scene measurements Scene: a 3-D function, g(x,y,z)
Review on Graphics Basics. Outline Polygon rendering pipeline Affine transformations Projective transformations Lighting and shading From vertices to.
Student Name: Honghao Chen Supervisor: Dr Jimmy Li Co-Supervisor: Dr Sherry Randhawa.
FREE-VIEW WATERMARKING FOR FREE VIEW TELEVISION Alper Koz, Cevahir Çığla and A.Aydın Alatan.
A Tutorial on using SIFT Presented by Jimmy Huff (Slightly modified by Josiah Yoder for Winter )
Computer vision: models, learning and inference M Ahad Multiple Cameras
Determining 3D Structure and Motion of Man-made Objects from Corners.
From cortical anisotropy to failures of 3-D shape constancy Qasim Zaidi Elias H. Cohen State University of New York College of Optometry.
Image features and properties. Image content representation The simplest representation of an image pattern is to list image pixels, one after the other.
Date of download: 6/1/2016 Copyright © 2016 SPIE. All rights reserved. (a) Optical image of fore and hind wings from a male S. charonda butterfly at different.
Shadow Detection in Remotely Sensed Images Based on Self-Adaptive Feature Selection Jiahang Liu, Tao Fang, and Deren Li IEEE TRANSACTIONS ON GEOSCIENCE.
Compressive Coded Aperture Video Reconstruction
outline Two region based shape analysis approach
Presented by :- Vishal Vijayshankar Mishra
Aim of the project Take your image Submit it to the search engine
CIS 4350 Image ENHANCEMENT SPATIAL DOMAIN
Recognition and Matching based on local invariant features
Presentation transcript:

Using eigencolor normalization for illumination-invariant color object recognition Speaker: 鄭雅勻 Date:2010/12/30 Zhenyong Lin, Junxian Wang, Kai-Kuang Ma Pattern Recognition 35 (2002) 2629 – 2642

Outline Introduction Related Work Implementation Results Conclusion

Introduction Color is one of salient features for color object recognition, however, the colors of object images sensitively depend on scene illumination. Color indexing, a histogram intersection algorithm to compare an observed histogram with those established from database. A color histogram is independent of common imaging conditions, such as orientation of a scene, absence or occlusion of colors but not the color of scene illumination changes. To overcome the lighting dependency problem, a color constancy or color normalization method can be used as a pre-processing step.

Introduction Color constancy process attempts to estimate the illumination, and the image colors are then corrected based on the estimated illumination to remove color bias and only the inherent colors are used for recognition. Calculate color-invariant features from images and use these features for indexing.

Related Work The failure of the color constancy method implies that the actual illumination of the scene might be different from the measured illumination B.V. Funt, L. Martin[1998] An approach called color constant color indexing, to demonstrate that the ratio of adjacent colors is relatively insensitive to illumination changes Funt and Finlayson [1995] Healey and Slater [1994] derived the functions of color distribution moments that are invariant to illumination changes. This approach does not attempt to recover the true colors of objects, but extracts the color-invariant features. To retain the image representation while keeping good indexing, a method called comprehensive color image normalization G.D. Finlayson, G.Y. Tian[1999]

Objective We present a new illumination-invariant color normalization algorithm, called eigencolor normalization. △ moments of color distributions [4] G. Healey, D. Slater, △ the normalization algorithm for planar patterns [6,7] J.G. Leu, S.C. Pei, C.N. Lin, The normalized color histograms of an object under different illuminations will become very similar to each other after the normalization process. The color object recognition can be performed more accurately by the color indexing on the normalized histograms.

Implementation The compact color image It is well-known that the colors of the same object under different illuminations are affine transformations to each other [4,5]. We can establish their affine-transformation relation as follows: Let and denote two n-dimensional histograms that represent the distributions of all color values and respectively.

Let H(R; G; B) be a color histogram; and the R,G and B values are quantized into 8 bits each (i.e.;0-255 discrete levels) The probability density function h(R; G; B) can be formed as Implementation

For the color histogram H(R; G; B), the central moment with order of k + r + l is denoted by and defined as Implementation

If we can find the affine transformation matrices transform Two histogram second-order central moment matrices the same object under two different illuminations direct histogram intersection matching X not good result Then, the affine-transformed Histograms will less correlated and more compact. O improved matching results

shape compacting technique that has been used for 2-D planar shape object normalization to compact 3-D color image. [6,7] The resulted image is called compact color image. Implementation ‧ it shows that a color image can be compacted by changing its coordinate system. the compact color image is variant to other non-affined transforms ex: skew transformation which is often caused by uneven lighting condition or curved object surface.

To make the color image more illumination-invariant, we need to further normalize the compact color image. Implementation matrix Q is orthogonal (proof)

If Q is orthogonal then that is Matrix Q is orthogonal From Eq.(14) From Eq.(9) From Eq.(8) (9)

Corollary 1. If the affine matrix is orthogonal; then. In this case, the compact color image is the same as the normalized color image. Proof. If the affine matrix is orthogonal Implementation the compact color image is the same as the normalized color image

Since every 3×3 orthogonal matrix is a rotation matrix in 3-D space [8], the normalized color image can be obtained by rotating the color of the compact color image. Implementation is the norm of the plane containing the main diagonal vector of the 3-D color space and the principal axis of the color distribution of the compact color image.

‧ Given the rotation axis and the rotation angle (according to the Rodrigues formula [8]), the rigid body rotation matrix can be estimated by : Rodrigues rotation formula Rodrigues’ rotation formula is an efficient algorithm for rotating a vector in space, given an axis and angle of rotation. By extension, this can be used to transform all three basis vectors to compute a rotation matrix from an axis-angle representation. the Rodrigues formula is: Implementation

By selecting Q = in our proposed eigencolor normalization algorithm, the new color coordinate system, called

Result A demonstration of eigencolor normalized histogram Here compare the histograms of a simple color image before and after exploiting the proposed eigencolor normalization processing. Sony DXC-950 3CCD color video camera under a white light source and interfaced to a Matrox Meter frame grabber card. Fig. 2. Color histograms before (from (a) to (c)) and after (from (d) to (f)) applying our eigencolor normalization processing. a,b,c  original R,G,B d,e,f  normalized R,G,B According to the algorithm, size of 45 × 45 is about 2 seconds on a Pentium III 500 Hz PC.

Result Color object recognition tests without illumination The color image database contains 66 model images with size of128 × 128 each Fig. 3 the monochrome version of model images The recognition or matching performance of each algorithm is the histogram distance.

Result All test images except T17 are correctly recognized as the best match. Further note that test image T17 is a much zoom-in and rotation version of model image 40 in Fig. 3.

Result ‧ Color object recognition tests with illumination changes a color image was captured by the same hardware but under different illuminations, white, red, green and blue.

Result ‧ Color object recognition performance comparison Sony DXC CDD video camera was used with gamma correction of and color temperature being set at 3200 K. region-of-interest image patch 40 × 40 each under four different illuminations Fig. 8. Histograms of two color images (a) and (d) are original histograms (b) and (c) are compact histograms (c) and (f) are eigencolor histograms These color images were taken under four different illuminations (from top to bottom in each sub-figure) Macbeth 5000 tube, Sylvania cool-white fluorescent tube, Phillips ultralume fluorescent tube, Sylvania 75 halogen bulb

Result two phases, an off-line training phase -> the eigencolor normalized histograms of the database images are generated an on-line matching phase-> the histogram of the eigencolor normalized image of a test object is first obtained ‧ On-line color object recognition under illumination changes syl-cwf -> Sylvania cool-white fluorescent tube ph-ulm -> Philips ultralume fluorescent tube halogen ->Sylvania 75 W halogen tube

Result

Conclusions In this paper, we present an effective way to normalize color images for correcting illumination changes, and consequently, improving color object recognition accuracy. The normalized color space, called eigencolor space, is aimed to be more invariant to various illumination changes, which mathematically corresponds to affine transformations as well as non-affine transformation from the original images. Results clearly show that our eigencolor representation approach outperforms in facilitating more accurate recognition of color objects under various illuminations.