An Image Retrieval Method Based on Hu Invariant Moment and Improved Annular Histogram

Similar documents
3D Model Retrieval Method Based on Sample Prediction

Stone Images Retrieval Based on Color Histogram

A Novel Feature Extraction Algorithm for Haar Local Binary Pattern Texture Based on Human Vision System

Harris Corner Detection Algorithm at Sub-pixel Level and Its Application Yuanfeng Han a, Peijiang Chen b * and Tian Meng c

New HSL Distance Based Colour Clustering Algorithm

Auto-recognition Method for Pointer-type Meter Based on Binocular Vision

RESEARCH ON AUTOMATIC INSPECTION TECHNIQUE OF REAL-TIME RADIOGRAPHY FOR TURBINE-BLADE

New Fuzzy Color Clustering Algorithm Based on hsl Similarity

World Scientific Research Journal (WSRJ) ISSN: Research on Fresnel Lens Optical Receiving Antenna in Indoor Visible

Title: Robust Registration of Multimodal Remote Sensing Images Based on Structural Similarity.

Performance Plus Software Parameter Definitions

Neuro Fuzzy Model for Human Face Expression Recognition

Accuracy Improvement in Camera Calibration

Image Segmentation EEE 508

IMP: Superposer Integrated Morphometrics Package Superposition Tool

Optimized Aperiodic Concentric Ring Arrays

Analysis of Server Resource Consumption of Meteorological Satellite Application System Based on Contour Curve

High-Speed Recognition Algorithm Based on BRISK and Saliency Detection for Aerial Images

Text Feature Selection based on Feature Dispersion Degree and Feature Concentration Degree

Euclidean Distance Based Feature Selection for Fault Detection Prediction Model in Semiconductor Manufacturing Process

Improving Template Based Spike Detection

x x 2 x Iput layer = quatity of classificatio mode X T = traspositio matrix The core of such coditioal probability estimatig method is calculatig the

Pattern Recognition Systems Lab 1 Least Mean Squares

Application of Decision Tree and Support Vector Machine for Inspecting Bubble Defects on LED Sealing Glue Images

DETECTION OF LANDSLIDE BLOCK BOUNDARIES BY MEANS OF AN AFFINE COORDINATE TRANSFORMATION

Mobile terminal 3D image reconstruction program development based on Android Lin Qinhua

Cubic Polynomial Curves with a Shape Parameter

A New Morphological 3D Shape Decomposition: Grayscale Interframe Interpolation Method

Web Text Feature Extraction with Particle Swarm Optimization

Low Complexity H.265/HEVC Coding Unit Size Decision for a Videoconferencing System

Journal of Chemical and Pharmaceutical Research, 2013, 5(12): Research Article

Study of Image Retrieval Method Based on Salient Points and Comprehensive Characteristics

Chapter 3 Classification of FFT Processor Algorithms

An Efficient Image Rectification Method for Parallel Multi-Camera Arrangement

Optimization for framework design of new product introduction management system Ma Ying, Wu Hongcui

Evaluation scheme for Tracking in AMI

VALIDATING DIRECTIONAL EDGE-BASED IMAGE FEATURE REPRESENTATIONS IN FACE RECOGNITION BY SPATIAL CORRELATION-BASED CLUSTERING

Fragments based tracking with adaptive multi-cue integration

( n+1 2 ) , position=(7+1)/2 =4,(median is observation #4) Median=10lb

A SOFTWARE MODEL FOR THE MULTILAYER PERCEPTRON

Dynamic Programming and Curve Fitting Based Road Boundary Detection

*Corresponding author. Keywords: Power quality, Assessment system, Harmonic evaluation, Comprehensive evaluation.

Evaluation of the Software Industry Competitiveness in Jilin Province Based on Factor Analysis

We are IntechOpen, the first native scientific publisher of Open Access books. International authors and editors. Our authors are among the TOP 1%

Normal Distributions

Analysis of Documents Clustering Using Sampled Agglomerative Technique

APPLICATION NOTE PACE1750AE BUILT-IN FUNCTIONS

Dimension Reduction and Manifold Learning. Xin Zhang

Diego Nehab. n A Transformation For Extracting New Descriptors of Shape. n Locus of points equidistant from contour

Outline. Research Definition. Motivation. Foundation of Reverse Engineering. Dynamic Analysis and Design Pattern Detection in Java Programs

A Flexible Hierarchical Classification Algorithm for Content Based Image Retrieval

Python Programming: An Introduction to Computer Science

Creating Exact Bezier Representations of CST Shapes. David D. Marshall. California Polytechnic State University, San Luis Obispo, CA , USA

On (K t e)-saturated Graphs

Administrative UNSUPERVISED LEARNING. Unsupervised learning. Supervised learning 11/25/13. Final project. No office hours today

Ajinkya P.Nilawar. M.E.(EC) Aurangabad,Maharashtra,India I. INTRODUCTION. Content Based Image Retrieval (CBIR)

The identification of key quality characteristics based on FAHP

Data Analysis. Concepts and Techniques. Chapter 2. Chapter 2: Getting to Know Your Data. Data Objects and Attribute Types

Study on effective detection method for specific data of large database LI Jin-feng

Chapter 1. Introduction to Computers and C++ Programming. Copyright 2015 Pearson Education, Ltd.. All rights reserved.

Automated Extraction of Urban Trees from Mobile LiDAR Point Clouds

An Efficient Algorithm for Graph Bisection of Triangularizations

On the Accuracy of Vector Metrics for Quality Assessment in Image Filtering

Fast Fourier Transform (FFT) Algorithms

An Improved Shuffled Frog-Leaping Algorithm for Knapsack Problem

DATA MINING II - 1DL460

are two specific neighboring points, F( x, y)

The Counterchanged Crossed Cube Interconnection Network and Its Topology Properties

An Efficient Algorithm for Graph Bisection of Triangularizations

Medical Image Segmentation of Blood Vessels Based on Clifford Algebra and Voronoi Diagram

2D Isogeometric Shape Optimization considering both control point positions and weights as design variables

Sectio 4, a prototype project of settig field weight with AHP method is developed ad the experimetal results are aalyzed. Fially, we coclude our work

Fundamentals of Media Processing. Shin'ichi Satoh Kazuya Kodama Hiroshi Mo Duy-Dinh Le

The Closest Line to a Data Set in the Plane. David Gurney Southeastern Louisiana University Hammond, Louisiana

COMPLEMENTARY SIMILARITY MEASURE

Handwriting Stroke Extraction Using a New XYTC Transform

A Comparative Study of Color Edge Detection Techniques

Bayesian approach to reliability modelling for a probability of failure on demand parameter

Efficient Eye Location for Biomedical Imaging using Two-level Classifier Scheme

Detection and Classification of Apple Fruit Diseases using Complete Local Binary Patterns

Evaluation of Support Vector Machine Kernels for Detecting Network Anomalies

Intermediate Statistics

A Study on the Performance of Cholesky-Factorization using MPI

Bezier curves. Figure 2 shows cubic Bezier curves for various control points. In a Bezier curve, only

. Written in factored form it is easy to see that the roots are 2, 2, i,

Image Fusion Based on Integer Lifting Wavelet Transform

EFFECT OF QUERY FORMATION ON WEB SEARCH ENGINE RESULTS

Size and Shape Parameters

Eigenimages. Digital Image Processing: Bernd Girod, Stanford University -- Eigenimages 1

Position and Velocity Estimation by Ultrasonic Sensor

Using a Dynamic Interval Type-2 Fuzzy Interpolation Method to Improve Modeless Robots Calibrations

Fast algorithm for skew detection. Adnan Amin, Stephen Fischer, Tony Parkinson, and Ricky Shiu

MATHEMATICAL METHODS OF ANALYSIS AND EXPERIMENTAL DATA PROCESSING (Or Methods of Curve Fitting)

Improvement of the Orthogonal Code Convolution Capabilities Using FPGA Implementation

Effect of control points distribution on the orthorectification accuracy of an Ikonos II image through rational polynomial functions

ENGI 4421 Probability and Statistics Faculty of Engineering and Applied Science Problem Set 1 Descriptive Statistics

3D MODELING OF STRUCTURES USING BREAK-LINES AND CORNERS IN 3D POINT CLOWD DATA

Criterion in selecting the clustering algorithm in Radial Basis Functional Link Nets

An Algorithm of Mobile Robot Node Location Based on Wireless Sensor Network

BASED ON ITERATIVE ERROR-CORRECTION

Transcription:

http://dx.doi.org/10.5755/j01.eee.0.4.6888 ELEKTROIKA IR ELEKTROTECHIKA ISS 139 115 VOL. 0 O. 4 014 A Image Retrieval Method Based o Hu Ivariat Momet ad Improved Aular Histogram F. Xiag 1 H. Yog 1 S. Dada 1 Z. Jiexia 1 1 School of Software achag Hagkog Uiversity Jiagxi achag 330063 Chia fxfb163@163.com 1 Abstract Image retrieval methods based o aular histogram of feature poits are calculatio efficiet ivariat to image rotatio ad traslatio trasform. However these methods have two mai disadvatages. Oe is that the aular histogram ca t describe the spatial distributio of feature poits accurately thus differet images may have similar aular histogram. Aother oe is the methods based o feature poits ca ot describe the shape iformatio of object. These disadvatages affect the retrieval accuracy to a certai extet. I this paper a image retrieval method based o Hu ivariat momets ad improved aular histogram is proposed. Firstly edge of image is detected ad feature poits are calculated based o the edge curvature. The image features are described based o both the edge ad poits. Aular histogram combied with stadard deviatio ellipse method is used to describe the spatial distributio of feature poits. Hu ivariat momet of the edge is used to represet the object s shape iformatio. Fially the similarity is measured based o both the poit feature ad the shape feature. Experimet results show that the proposed method ca improve the image retrieval precisio effectively. Idex Terms Image retrieval Hu ivariat aular histogram shape iformatio. I. ITRODUCTIO momet With the developmet of computer techique ad image processig techique cotet-based image retrieval (CBIR) has become a active ad fast-advacig research area i image retrieval. Poit feature is widely used to CBIR because the feature poits of image cotai abudat iformatio ca describe may details of image what s more retrieval method base o feature poits is calculatio efficiet. A image retrieval method based o feature poits average momets is proposed i [1]. However for iformatio represeted by sigle feature is limited may researchers combied differet features to improve the retrieval performace. I [] two kids of histogram are used to describe image feature. Oe is that the histogram of circumscribed circle radius from adjacet edge to the cetroid of feature poits aother is the histogram of distaces betwee feature poits ad the top left corer of the image. I [3] ad [4] the authors use the covex Mauscript received March 8 013; accepted August 3 013. This research was fuded by grat (o. 01BBF6007) from the Sciece ad Techology Project of Jiagxi Provice ad grat (o. 013BAB1101) from the Youth Sciece Foudatio of Jiagxi provice. This research was performed i cooperatio with the Istitutios. 67 hull of iterest poits to describe the image feature combie SVM learig method ad Zerike momets respectively to improve the efficiecy of image retrieval moreover both methods use aular histogram to improve the performace of retrieval. Sice the method based o aular histogram is calculatio efficiet ivariat to image rotatio ad traslatio ad it ca idicate the spatial distributio of feature poits to a certai extet. However the methods based o aular histogram caot describe the spatial distributio of feature poits accurately thus differet images may have similar aular histogram ad the methods based o feature poits caot describe the shape iformatio of object. I [5] a image shape characteristics extractig algorithm based o regioal pseudo Zerike momet is proposed the used to image retrieval. I [6] cotour poits are used to image retrieval. The cotour poits possess the shape characteristics ad poit features at the same time. I this paper cosiderig the fact that limited iformatio represeted by sigle feature is isufficiet to describe the image features a image retrieval method based o Hu ivariat momets ad improved aular histogram is proposed. Firstly the edge feature of image is detected by edge operator ad the the feature poits are extracted based o the edge curvature. Secodly image features are described based o the edge ad poits. I order to overcome the disadvatage of method based o aular histogram stadard deviatio ellipse [7] method is used to describe the spatial distributio of feature poits accurately. At the same time Hu ivariat momets [8] of feature edge are used to describe the shape characteristic. Fially similarity betwee images are measured based o these two features. The improved aular histogram method takes more cosideratio of feature poits geometry space distributio combied with shape feature the retrieval accuracy is expected to be improved. Experimet results show that this algorithm has good retrieval results. II. THE PROPOSED ALGORITHM The block diagram of the proposed method is show i Fig. 1. First of all the feature edge ad poits of query image ad curret image i image database are detected. The the space distributio of feature poits is described based o improved aular histogram method ad shape feature is

ELEKTROIKA IR ELEKTROTECHIKA ISS 139 115 VOL. 0 O. 4 014 represeted based o Hu ivariat momet of edge feature. Fially similarity of images is measured based o these two features ad the image retrieval results are obtaied. I order to retrieve image accurately ad quickly effective feature descriptio method is ecessary. After feature poit detectio aular histogram method is adopted to describe the poit feature. The poits are divided ito a series of cocetric rig regios with equal iterval. The color feature of poits i every cocetric rig regio is described by histogram statistics. The stadard deviatio ellipse method is used to describe the geometric distributio iformatio of poits iclude distributio ceter distributio rage ad distributio directio. For the reaso that the methods based o feature poits caot describe the shape iformatio of object. Hu ivariat momets of edge are used to describe the shape feature i the ext step. Edge feature ca represet the shape characteristics of objects ad Hu ivariat momets are robust to rotatio scale ad traslatio trasformatio. Fig. 1. Block diagram of the proposed method. A. Feature Extractio There exist may classical edge detectors such as Cay operator Roberts operator Prewitt operator Sobel operator Kirsch operator Compass operator ad so o. Cay operator [9] is adopted i this paper because it has the abilities of preservig importat edge iformatio i image producig little false edge havig good ati-oise ad edge locatio performace. For the origial image show i Fig. the edge detected by Cay operator is give i Fig.. There are two commo ways to feature poit extractio. Oe is that based o gray of image select poits where the gray iformatio chages obviously. Harris is the commo used poit of iterest detectio algorithm based o gray iformatio of image. Aother way is based o the edge image the poits where the edge curvature chages greatly are selected as feature poits. The mai drawback of methods base o gray iformatio is susceptible to oise. Moreover large amout of feature poits located i the regios of backgroud may be cotaied. These poits may affect the performace of image retrieval because most of them are disturbig iformatio or iformatio that people ot cocer. The feature poits extractig method based o edge curvature has good ati-oise performace ca reduce the useless poits located i backgroud or oise poits to a great extet by selectig poits o the edge where the curvature chages greatly. Figure (c) shows the poits detected by the poit of iterest detectio algorithm [10] based o the edge image of Fig.. The compariso results are give i Fig. 3. Where the testig results obtaied by Harris based o gray image are show i Fig. 3 ad feature poits extracted based o edge image are show i Fig. 3. It ca be see that compared with Fig. 3 there are more poits located i the backgroud regio are cotaied i Fig. 3. I order to improve the accuracy of image retrieval the iterest poit detectio algorithm [9] based o local curvature of edge is adopted i this paper. (c) Fig.. Feature detectio results. The origial image; Edge detected by Cay operator; (c) Feature poits extracted based o edge curvature i. Fig. 3. Feature poits detectio results compariso. Feature poits detected by Harris based o grey image; Feature poits extracted based o edge image. B. Shape Feature Descriptio based o Hu Momets Hu ivariat momets also called geometric ivariat momets are widely used i image processig because of the robustess to image traslatio scale ad rotatio trasformatio. I this paper the 7 order Hu ivariat momets are calculated to describe the shape feature based o the ormalized cetral momets of the edge image. Firstly for image f ( x y) with size M the p q order momets are defied as mpq M p q x y f ( x y). m1 1 (1) 68

ELEKTROIKA IR ELEKTROTECHIKA ISS 139 115 VOL. 0 O. 4 014 The the cetral momets are defied as M p q pq ( x x ') ( y y ') f ( x y) m1 1 () where x' m10 m00 y ' m01 m00. I order to ormalize the cetral momet pq each pq divided by as (3) 00 where ( p q) / 1.. The ormalized cetral momet is show pq pq / ( 00 ) (3) Fially the 7 order Hu ivariat momets are defied as follows: M 1= 0 + 0 M ( 0 0) 4 11 M3 ( 30 3 1 ) (3 1 03) M 4 ( 30 1) ( 1 03) M 5 ( 30 3 1 )( 30 1)(( 30 1) 3( 1 03) ) (3 1 03)( 1 03)(3( 30 1) ( 1 03) ) M6 ( 0 0)(( 30 1) ( 1 03) ) 4 11 ( 30 1)( 1 03) M 7 (3 1 03)( 30 1)(( 30 1) 3( 1 03) ) ( 30 3 1 )( 1 03 )(3( 30 1 ) ( 1 03 ) ). C. Spatial Distributio of Poits Feature Descriptio based o Improved Aular Histogram At the begiig the image is divided ito k cocetric rig regios based o the distributio of feature poits. The cocetric rig method used i this paper is similar to that of i [11] which has good traslatio scale ad rotatio ivariace. Firstly the maximum distace betwee the features poits i the curret image is calculated ad makes it as the maximum diameter. The divide the feature poits ito k cocetric rig regios with equal itervals. This kid of image partitio method has traslatio scale ad rotatio ivariace ad ca be used to improve the robustess of image retrieval to image trasformatio. Obviously the larger the k is the more clear the image detail iformatio represetatio. Fig. 4 shows the image divided ito cocetric rig regios based o the distributio of feature poits i Fig. (c). After partitio the image ito k aular blocks the color feature of poits i every cocetric rig regio is described by histogram. Suppose hm value of the m-th cocetric rig. It has where m is (4) meas histogram statistical hm m (5) is the sum of the color value for poits i the m-th cocetric rig ad is the sum of the color value for poits i the whole image. Aular histogram method ca describe the distributio desity of feature poits but it is icompetet to the descriptio of spatial distributio ad differet images may have similar histogram. It is foud that stadard deviatio ellipse ca describe the spatial iformatio of feature poits i -dimesioal space. The spatial distributio iformatio described by stadard deviatio ellipse icludes the distributio directio ad the distributio shape of feature poits. The descriptio method based o stadard deviatio ellipse is robust to traslatio ad rotatio trasformatio of images. Therefore as a supplemet the stadard deviatio ellipse method is adopted to describe the spatial distributio of feature poits. The descriptio ad defiitio of the stadard deviatio ellipse iclude maily three parameters the major axis L l the mior axis Ls ad the mai directio i.e. the directio of major axis: ( xi x)cos ( yi y)si i1 Ll l ( xi x)si ( yi y)cos i1 Ls s where l s are the stadard deviatio of the major axis ad the mior axis of the ellipse they equal to half legth of major axis ad mior axis respectively; {( xi yi ) i 1 } is the coordiate values of feature poits: x xi / i1 y yi / are the mea ceter coordiates i x directio i1 ad y directio for all poits; the value rage of is from 0 to. A stadard deviatio ellipse computed based o the distributio of feature poits i Fig. (c) is show i Fig. 5 ( xi x) ( yi y) i1 i1 arcta ( xi x)( yi y) i1 ( xi x) ( yi y) 4( ( xi x)( yi y)). i1 i1 i1 (8) A oe dimesioal feature vector is costructed based o the aular histogram ad the stadard deviatio ellipse as (9) where i order to reduce the computatioal complexity the aular histogram is ormalized. (6) (7) h[1] h[] h[ k] Ls H. (9) k k k Ll h( m) h( m) h( m) i1 i1 i1 69

ELEKTROIKA IR ELEKTROTECHIKA ISS 139 115 VOL. 0 O. 4 014 Fig. 4. The image is divided ito cocetric rig regios. Fig. 5. Stadard deviatio ellipse for feature poits i Fig. (c). III. SIMILARITY MEASUREMET The similarities of spatial distributio ad shape feature are measured based o Euclidea distace as (10) ad (11) k 1 p = ( O ( ) I ( )) i1 S H i H i (10) where HO ( i) ad HI ( i) are the i-th feature vector defied by (9) for the query image O ad the image for matchig I respectively 7 S s = ( ( SO[ j] SI[ j]) ) j1 (11) where SO[ j] is the j-th feature vector of Hu ivariat momets of the query image O SI [ j] is the j-th feature vector of Hu ivariats momets of image I. Equatio (10) ad (11) express the spatial distributio of poit feature ad shape feature respectively. The fial similarity measuremet formula is educed by weightig the poit ad shape features as follows S= S S p s (1) where ad are parameters satisfy 1. The smaller the value S is the more similar the two images are. IV. EXPERIMETAL RESULTS AD AALYSIS I order to demostrate the performace of the proposed method it is compared to the method based o aual histogram ad the method based o poit of iterest preseted i [1] where the o iterest poits located i the backgroud regio were elimiated based o local Zerike ivariat momets ad the similarity is measured based o spatial cohesio of iterest poits. (c) Fig. 6. A example of image retrieval results for three methods. The proposed method; The method based o aual histogram; (c) The method i [1]. I the experimets 1000 images from Corel image database are selected. These images are divided ito 10 categories ad each category cotais 100 images. For the diosaur image i Fig. a example of image retrieval results with three methods are show i Fig. 6. For each method the top 30 most similar images are retured automatically. It ca be see that the proposed method is the most effective amog three methods there are 3 19 ad 16 correct images retrieved for the proposed method the method based o aual histogram ad the method i [1] respectively. 70

ELEKTROIKA IR ELEKTROTECHIKA ISS 139 115 VOL. 0 O. 4 014 Test image TABLE I. COMPARISO OF PRECISIO RATE FOR THREE METHODS. P 10 (%) P 0 (%) P 30 (%) Proposed Aual histogram Method Aual Method Aual Method Proposed Proposed i [1] histogram i [1] histogram i [1] Africa people 0. 18.7 0 16.4 15.15 15 11.33 10.67 10 Beach 35.4 33.9 30 8.7 4.6 15 3.17 0.13 13.3 Bulidigs 33.1 3.5 0 5.4.3 15 33.5 8.0 13.3 Buses 34.7 30.6 30 0.5 18.9 0 14.4 13.67 13.3 Diosaurs 74. 73.8 60 66.3 6.5 40 64.7 6.0 40 Elephats 41.3 40.4 40 3.8 0.0 0.47 19.67 0 Flowers 75.8 7.3 70 56.35 55.05 55 43.5 4.4 40 Horses 50. 49.6 50 49. 48.75 60 46. 45.67 46.7 Moutais 55.6 54.7 40 50.75 46.3 30 48.33 40.67 3.3 Food 15.3 1.8 10 1.6 1.3 5 9.8 7.5 3.3 The commoly used performace measuremet precisio rate is used for the evaluatio of retrieval performace. Precisio rate is defied as follows R = C P R (13) where C is the umber of images which belog to the same category with query image O ; R is the total umber of images which output automatically by the retrieval method. Precisio P R measures the accuracy of the retrieval. The experimet strategies are as follows: 10 query images are selected radomly from each category. P 10 P0 ad P 30 are calculated respectively for each query image based o (13). For 10 query images belog to the same category the average values P 10 P0 ad P 30 are computed. The compariso of precisio rate is show i Table I. It ca be see from Table I there are some easy categories such as Diosaurs ad Flowers these images have clear object ad simple backgroud o which all the algorithms perform well. Ad there are some hard categories such as Africa people ad Food the object i these images is ot clear ad the backgroud is complicated o which all the algorithms perform poorly. For the images of elephats buses ad horses the results of the proposed method are almost the same as those of the other two methods. That is because the backgroud of the image is very complex ad the object i the image is smaller tha the backgroud regio there are large portio of iterest poits i the backgroud regio still preserved whe measurig the similarity ad these poits will disturb the poits i the iterest object. For the images of beach buildigs ad moutais the accuracy of the proposed method is better tha the other two methods. That is because the whole image is retrieval object for these images the spatial distributio of poits ad the shape of object ca be well described by the proposed method. By cotrast the method based o aual histogram ca either describe the spatial distributio of feature poits accurately or describe the shape iformatio of object the retrieval accuracy is poorad for the method based o poit of iterest preseted i [1] [13] where the o iterest poits located i the backgroud regio were elimiated based o local iformatio of poits the algorithm performs poorly because there is o backgroud or obvious object i these images that algorithm degeerate ito measurig the similarity based o spatial cohesio of iterest poits directly at the same time it did ot use the shape iformatio of image either. V. COCLUSIOS Poit feature cotais may detail iformatio of image has become a importat visual feature i image processig field. Edge feature is widely researched because it ca describe the shape characteristic of target. A cotet-based image retrieval method is proposed i this paper which takes both poit feature ad edge feature as clue. The aual histogram combied with stadard deviatio ellipse is used to describe the poit feature ad the Hu ivariat momets of edge are used to describe the shape feature. Experimetal results show that this retrieval method has better performace tha the classical methods based o iterest poits. REFERECES [1] Huag Yuayua Liu igzhog Trademark retrieval based o feature poits average momets Joural of image ad graphics vol. 115 o. 4 pp. 637 644 010. [] Qi Heg Li Keqiu She Yamig Qu Weyu A effective solutio for trademark image retrieval by combiig shape descriptio ad feature matchig Patter Recogitio vol. 43 o. 6 pp. 017 07 010. [Olie]. Available: http://dx.doi.org/10.1016/j.patcog. 010.01.007 [3] Su Xiaohog Dig Ji Ma Peiju Image retrieval by covex hulls of iterest poits ad SVM-based weighted feedback Chiese Joural of Computers vol. 3 o. 11 pp. 1 8 009. [4] Meg Fajie Guo Baolog Li Xiwei Guo Lei ew image retrieval method based o covex hull of iterest poits Joural of Optoelectroics Laser vol. 1 o. 6 pp. 936 939 010. [5] S. Li M. C. Lee C. M. Pu Complex Zerike momets features for shape-based image retrieval IEEE J. Tras. Systems Ma ad Cyberetics vol. 39 o. 1 pp. 7 37 009. [Olie]. Available: http://dx.doi.org/10.1109/tsmca.008.007988 [6] Shu Xi Wu Xiaoju Pa Lei A ovel approach to image retrieval based o distributio of shape cotour s poits Joural of Optoelectroics Laser vol. 0 o. 10 pp. 1385 1389 009. [7] Liu Tao Du Qigyu Ya Haowe Spatial similarity assessmet of poit clusters Geomatics ad iformatio Sciece of Wuha Uiversity vol. 36 o. 10 pp. 1149 1153 011. [8] Gao Yuefag Ha Guoqiag Li Guiqig Wo Ya Wag Dog Developmet of curret momet techiques i image aalysis Joural of image ad graphics vol. 14 o. 8 pp. 1494 150 009. [9] Li Wei She Zhekag Li Biao Corer detectio based o local curvature fuctio Compute egieerig ad desig vol. 8 o. 11 pp. 595 596 007. [10] J. Cay A computatioal approach to edge detectio IEEE J. Tras. Patter Aalysis ad Machie Itelligece vol. 8 o. 6 pp. 679 698 1986. [Olie]. Available: http://dx.doi.org/10.1109/ TPAMI.1986.4767851 [11] Guiguag Dig Lu Zhag Xiaoxu Li Video aotatio based o adaptive aular spatial partitio scheme IEICE J. Electroics Express o. 7 pp. 7 1 010. [1] Fu Xiag Zeg Jiexia A ovel image retrieval method based o iterest poits matchig ad distributio Chiese Joural of laser vol. 37 o. 3 pp. 774 778 010. [13] Tafeg Su Xighao Jiag Chegmig Jiag Yaqig Li A Video Cotet Classificatio Algorithm Applyig to Huma Actio Recogitio Elektroika ir Elektrotechika (Electroics ad Electrical Egieerig) vol. 19 o. 4 pp. 61 64 009. [Olie]. Available: http://dx.doi.org/10.5755/j01.eee.19.4.4056 71