MATEC Web of Conferences 22, 0102 0 ( 2015) DOI: 10.1051/ matecconf/ 2015220102 0 C Owned by the authors, publshed by EDP Scences, 2015 Research on Categorzaton of Anmaton Effect Based on Data Mnng Na N Department of Anmaton, Art College, Henan Unversty, Kafeng, Henan, Chna ABSTRACT: Nowadays, the producton process of anmaton effect s ncreasngly developed, and ts effect s also growng better. But n most cases, the categorzaton of specal effect added to the anmaton s confusng due to excessve varatons. Data mnng wll desrably solve the problem of anmaton effect categorzaton, so the applcaton of data mnng n the anmaton effect categorzaton becomes the hot spot n research and analyss at present. Ths artcle makes a detaled analyss on relevant algorthm of data mnng technology, that s, the k applcaton of averagng method, k central pont method and relatonal degree algorthm n problem of anmaton effect categorzaton. It provdes a clear method of categorzaton for anmaton effect. Thereafter, t also concludes the accuracy of anmaton effect categorzaton can be greatly mproved through reasonable algorthm ntegraton n the treatment of anmaton effect categorzaton by data mnng. Keywords: data mnng, anmaton effect categorzaton, cluster analyss, relatonal degree 1 INTRODUCTION In the background of openng up polcy, the establshment of socalst road wth Chnese features realzes rapd development for varous trades n Chna. The ncorporaton of dversfed elements and varous ndustres provdes unque vtalty for varous ndustres, and our Chnese producton process of anmaton effect also rapdly develops from mmaturty to excellence. We are always learnng and tryng to use the globally advanced new technologes. Prepare the whole anmaton effect based on thnkng and experence n prevous anmaton makng, dfferent eras, objectves and promnent effect. So there are many knds of anmaton effect wth dfferent advantageous functons. Therefore, how to dstngush a great number and many types of anmaton effect and provde obvous categorzaton ndex becomes the hot spot nowadays. In the current bg data era, many use the mathematc method to solve the actual problems, and the methods are ncreasngly perfected. Ths artcle tres to use data mnng method for detaled research and analyss on the problem of anmaton effect categorzaton. Fgure 1. Anmaton effect of herarchy sense Ths s an Open Access artcle dstrbuted under the terms of the Creatve Commons Attrbuton Lcense 4.0, whch permts unrestrcted use, dstrbuton, and reproducton n any medum, provded the orgnal work s properly cted. Artcle avalable at http://www.matec-conferences.org or http://dx.do.org/10.1051/matecconf/20152201020
MATEC Web of Conferences Fgure 2. Anmaton effect of fuzzy Fgure 3. Typcal flow chart of data mnng 01020-p.2
ICETA 2015 Fgure 4. Flow chart of data mnng for anmaton effect categorzaton 2 MODEL ESTABLISHMENT AND ITS SOLU- TION In the current socety wth developed mass meda, there are ncreasngly growng types of anmaton effect that are used to mprove the anmaton effect and have more comprehensve functons. For example, the anmaton wth many scenc spots wll requre specal effect to strengthen the sense of herarchy, and the anmaton requrng artstc concepton shall be added wth specal effect of local fuzzy or renderng; the acton scene wth vsual mpact requres mproved specal effect of textle, and so on. Fgure 1 and Fgure 2 add the specal effect of herarchy sense and fuzzy for the anmaton to provde satsfactory vsual effect for the anmaton. It can be seen from Fgure 1 and Fgure 2 that, despte the emphass of those two Fgures, n most cases the anmaton s composed by multple specal effects, so the specal effect categorzaton added for the anmaton becomes fuzzy and the accuracy wll be greatly reduced. Data mnng wll desrably solve the problem of anmaton effect categorzaton, so the applcaton of data mnng n the anmaton effect categorzaton becomes the hot spot n research and analyss at present. 2.1 Prncple of data mnng for anmaton effect categorzaton The data mnng refers to extract the hdden nformaton whch s hard to be dentfed by human efforts by mnng and collectng a large number of fuzzy data or partal data; then structure and screen out avalable nformaton therefrom; and deduct, generalze and make statstcs; then fnally vsualze those nformaton, and fnd out the relevance of those data by macro analyss and research; categorze them based on comprehensve consderaton of ther typcal features. Whle the data mnng algorthm used for treatment of anmaton effect categorzaton problem n ths artcle manly focuses on extractng and mnng the types and specal effect of producton process for each anmaton effect, generalzng and makng statstcs therefor, fndng out the relevance of varous anmaton effects and fnally categorze and dentfy them based on relevance of varous anmaton effects. Fgure 3 s the typcal flow chart of data mnng. It s known from the fgure 3 that the tradtonal data mnng method s to use varous data analyss algorthm based on the obtaned data, to establsh a model, to categorze and assess the obtaned data; predct the varables and fnally to concentrate and group accordng to the relevance. Whle for the treatment of anmaton effect categorzaton, t manly mnes and analyzes the 3D or mult-dmensonal data. Fgure 4 s the flow chart of data mnng for anmaton effect categorzaton. It can be seen from Fgure 4 that ths artcle makes pre-treatment for the anmaton effect; nvests t nto type and tranng concentraton; then comes nto feature selecton mode; conducts feature analyss for obtaned data; extracts feature data; categorzes the anmaton effect and nspects the categorzaton accu- 01020-p.3
MATEC Web of Conferences racy of obtaned result. Thus t wll be avoded to select the data wthout dfferences and wll make other serous mstakes. And the fnal nspecton of categorzaton accuracy for anmaton effect categorzaton s also an mportant ndex for assessment model accuracy, and t mposes a great mpact on the clarty and accuracy of anmaton effect. 2.2 Analyss on data mnng algorthm As mentoned above, the data mnng refers to extractng the hdden nformaton hard to be dentfed by human efforts by mnng and collectng a large number of fuzzy data or partal data, then structurng and screenng out avalable nformaton therefrom, and makng mathematcal statstcs, then analyzng the relevance of those data, and fnally categorzng them based on comprehensve consderaton of ther typcal features. Ths artcle uses three data mnng algorthms to conduct correspondng analyss and research respectvely on the problems of anmaton effect categorzaton. 2.2.1 Analyss on cluster algorthm The cluster analyss s a process that generalzes and categorzes the obtaned data accordng to ts smlarty, and put the smlar data nto one set. Then mark every obtaned data aggregate. Ths cluster algorthm analyss may solve varant problems and desrably fnd the relatve features for varous data aggregate. There are two types of data aggregate n the cluster analyss, whch are data matrx and adjacent value matrx. Data matrx, that s, the matrx of m n by representng n target wth m varables, shown as follows: x L x L x 11 1 f 1p L L L L L x L x L x L L L L L x L x L x 1 f p n1 nf np Whle adjacent value matrx refers to a matrx of n n, whch represents n matrx of data smlarty as follows: 0 d(2,1) 0 d(3,1) d(3, 2) 0 L L L 0 dn (,1) dn (,2) L L 0 Among whch, the smlarty measured between data and data j s d (, j ). As d (, j ) approaches 0, t ndcates the data and j are more smlar; smlarly, as d (, j ) value grows, t ndcates the data and j have less smlarty. So there may be errors between data aggregates n dfferent categorzatons. Whle the error calculaton method used n the cluster analyss of ths artcle s as follows: 1) The absolute error of mean: 1 S x m x m L x m n f 1f f 2f f nf f x,, 1 f L xnf represent n measured values for f, whle the average of f s m f 1 m x x L x n f 1f 2f nf, and 2) Standard measured value: The absolute error of mean proposed above can desrably reduce the mpact of outlers on calculaton result. That s to say, the measured value of outlers obtaned from absolute error of mean can be well dentfed. The calculaton formula wth measured value s as follows: z f x f m s f f The smlarty between data aggregates n cluster analyss s determned accordng to comparson of adjacent values for all data. Whle there are manly two calculaton methods for d (, j ): 1) Eucldean dstance: 1 1 2 2 L 2 2 2 d (, j) x xj x xj xn xjn ( x, x, L, x ) and 1 2 n j1 j2 jn j ( x, x, L, x ) n the formula above respectvely represents the data n two n -dmensonal data aggregate. 2) Manhattan dstance: d j x x x x x x (, ) 1 j1 2 j2 L n jn The Eucldean dstance and Manhattan dstance above shall meet the requrements below, namely: 01020-p.4
ICETA 2015 d (, j) 0 d (,) 0 d (, j) d( j,) d (, j) dh (, ) dhj (, ) The cluster algorthm analyss ntroduced n ths artcle solves the categorzaton problem of anmaton effect wth two methods, namely k averagng method and k central pont method. The k averagng method s to dvde n data nto k data aggregates by takng k as the parameter, and the categorzaton result s the hgher data smlarty n the data aggregate whch leads to hgher dfference between data aggregates. However, the smlarty of data aggregate s ts barycenter. It s defned as follows: k E pm 1 pc 2 E represents the sum of quadratc dfference of all data n the data aggregate, p the specfed data pont, and m the mean of data aggregate C. Randomly select k data n the data set D as the ntal center of data aggregate, judge each data accordng to the formula above through the obtaned mean, put t nto the data aggregate wth the most smlar features, change the mean of data aggregate, and calculate the mean of data untl t does not change. Therefore, the k averagng method wll have a better performance when the data aggregate separated from anmaton effect has a relatvely obvous dfference. Whle the k central pont method s to select a data n each data aggregate to represent the whole data aggregate, and cluster the rest of data nto the data aggregate of selected data wth the hghest smlarty n order to categorze the data of anmaton effect. It s defned as follows: k E po 1 pc j o j represents the data selected from C j. Now we randomly select k data from data set D as the ntal selected data; assgn the rest data nto the data aggregate represented by the selected data accordng to the above formula; select one more data o ; change o j nto o ; f the generally pad functon s S 0, a new data set wll be formed, and the repeated calculaton s requred untl t s unchanged. The result calculated by central pont s used to assess the mean of dssmlarty between selected and resdual data through the functon S, but there are four cases as to whether the data o can be changed nto o. j 01020-p.5
MATEC Web of Conferences Fgure 5. Four statuses to change non-representatve data As known from the fgure above, when p s n the data aggregate represented by o j, f o j s changed nto o, and p s closest to the resdual selected data o, p shall be put nto o ; when p s n the data aggregate represented by o j, f o j s changed nto o, and p s closest to the data o, p shall be put nto o ; when p s n the data aggregate represented by o, f o j s changed nto o, and p s closest to the data o, t shall not be subject to a second assgnment; when p s n the data aggregate represented by o, f o j s changed nto o, and p s closest to the data o, the p shall be put nto o ; every assgnment wll make the S functon of E change. If S 0, a new data set wll be formed; If S 0, t wll ndcate the change can be accepted, and thus t s not subject to another assgnment, that s to say the categorzaton s fnshed. 2.2.2 Analyss on relatonal degree algorthm Relatonal degree algorthm refers to an algorthm that can desrably fnd out the relatonshp and relatonal degree between obtaned data sets after obtanng the vald data through varous data mnng methods and algorthms. Such as Aa B, that s to say, when the feature value of A s a, the feature value of frequent tem set B wll have a hgher probablty to be b. So the relatonal degree algorthm may be used to detect the abnormal data and acheve hgh accuracy wth the help of other cluster algorthm. To fnd out relatonal degree, we shall frst obtan the frequent tem set and the relevance among the data. And the frequent tem set refers to elements wth hgh appearance rate n the data aggregate, and ts threshold scope s defned by support functon. Its functon s as follows: Support( A, B) P( AYB) Support A, B N( AB, ) Among whch, N s the total number of tmes, and N ( AB, ) the number of tmes the elements A and B appear smultaneously. As the mnmum support threshold Mn-sup s found, ths parameter may only capture the data larger than Mn-sup, so t wll be easy to obtan the frequent tem set n the objectve aggregate. As the frequent tem set s found, we may calculate the relatonal degree, that s to say we may get a relevance that has hgher relatonal degree and an mportant mpact on the analyss. When calculatng the relatonal degree, ts confdence level shall be met. The calculate formula s as follows: N confdence( A B) P( B A) 01020-p.6
ICETA 2015 Support confdence( A B) Support A, B A The formula above ndcates the confdence level of A B s the probablty of B under A con- B dton, namely the confdence level of A may be obtaned through calculaton of Support A, B and Support A, whch also ndcates A B can be easly calculated after determnng the value of Support A, B and Support A, that s to say we may obtan the correspondng relatonal degree among data and nspect ts accuracy after determnng the value of, Support A. Support A B and It s known from the analyss above that, n solvng the problem of anmaton effect categorzaton, the data mnng dentfes and categorzes the specal effect based on analyss of creaton purpose and process of anmaton effect, as well as through dscoverng and extractng the data of anmaton effect. Ths artcle categorzes the anmaton effect wth k averagng method and k central pont method and nspects the abnormal data of anmaton effect categorzaton by ncorporatng k central pont method and relatonal degree algorthm to mprove the categorzaton accuracy. 3 CONCLUSION Ths artcle wll frst analyze the development stuaton of current anmaton effect accordng to the detaled stuaton of socety, and then fnd out the man reason that causes the problem of anmaton effect categorzaton as well as the necessty to apply the data mnng theren. Ths artcle makes a detaled analyss on relevant algorthm of data mnng technology, namely the applcaton of k averagng method, k central pont method and relatonal degree algorthm n problem of anmaton effect categorzaton. It provdes a clear method of categorzaton for anmaton effect. In solvng the problem of anmaton effect categorzaton, ths artcle uses the data mnng to dentfy and categorze the specal effect based on analyss of creaton purpose and process of anmaton effect, as well as through dscoverng and extractng the data of anmaton effect. Ths artcle categorzes the anmaton effect wth k averagng method and k central pont method and nspects the abnormal data of anmaton effect categorzaton by ncorporatng k central pont method and relatonal degree algorthm to mprove the categorzaton accuracy. ACKNOWLEDGEMENT Ths work was supported by research of creatve thnk ng model tranng n art and desgn educaton of colle ges and unverstes n Hennan (142400411094). REFERENCES [1] Faber, L. & Walters, H. 2004. Anmaton Unlmted. Translated by Wang, K.Y. Shangha People's Fne Arts Publshng House. [2] Stephen Vsro. 2004. Dgtal Cartoon Art. GU, C. Shangha People's Fne Arts Publshng House. [3] Steve Capln, Adam, Banks. 2004. Dgtal Graphcs Art. Translated by Cao, T.Q. Shangha People s Fne Arts Publshng House. [4] Lu, F., Wu, P... Abnormal Data Capture and Analyss Based on Data Mnng. The research Chna nformaton system. [5] Q, H.Y., Wu, X.J., Wang, S. & Yang, J.Y. 2010. A Collaboratve FCPM Fuzzy Clusterng Algorthm, Pattern Recognton and Artfcal Intellgence, 23 (1): 120-122. [6] Gan, L. & Wang, X. 2008. Study on Cluster Analyss n Data Mnng. Coastal Enterprses and Scence And Technology, 42 (3): 24-26. [7] Lu, H.Y., Chen, J. &Chen, G.Q. 2002. Revew of Data Categorzaton Algorthms for Data Mnng. Journal of Tsnghua Unversty (NATURAL SCIENCE EDITION), 42 (06): 727-730. [8] Kong, L.F., Wang, H. & Lan, Q.S. 2006. Data Organzaton Method Based on Assocaton Rules Mnng. Computer Engneerng, 32 (21): 12-15. [9] Fu, Z.W... 2010. Research on Incremental Updatng Algorthm for Assocaton Rules. Journal of Software, 9 (8): 52-53. [10]L, J.F., Xao, R. & Qn, J. 2010. Research and Desgn of Incremental Mnng Assocaton Rules. Journal of Software, 9 (10): 77-79. 01020-p.7