Fig.2.Three original images(max_x×max_y).a 238×170.b 185×196.c 275×225
Fig.3.Templates for matching(temp_x×temp_y).a Template 1.b Template 2
3 Experiment and comparison
The genetic operations and GA parameters are presentedin Table 1 and Table 2.The fitness is defined as
temp_ytemp_x??j?0i?0f(x,y,rate,?)?temp(i,j)fitness?1.0?(temp_x)?(temp_y)?255
(4)
In Eq.4,temp(i,j) is the gray level of the coordinates(i,j)in the template image,the width and height of which aretemp_x and temp_y.f(x,y,rate,?)gives the gray level in theoriginal image,the coordinates of which are calculated bytranslation from(x,y),and by changing the scale and therotation angleθfrom the template.Since the images are
256gray-level images,in Eq.4,division by 255 ensures that theresulting fitness is between 0 and 1.The maximum numberof generations is limited to 300,and the threshold of thematching rate is set to 0.9.6That is to say,if within 300generations the matching rate can reach 0.9,then it is saidthat the template is found in the original image(the template matched the original image by thethreshold).Otherwise,the result gives the best match until the trainingreaches the 300thgeneration.
The results of GA-based face recognition are given inFig.6 and Table 3.Figure 6a,c and d are searched to matchthe template Fig.3a,while Fig.6b is matched to Fig.3b.Figure 6a and b reach the matching rate 0.9 within 300generations,while Fig.6c and d cannot reach the matchingrate 0.9 within 300 generations(the best match is given inTable 3).In the images in Fig.6a–c,we see that the resultgiven matches the template well.The coordinates
(x,y) ,therate of scale,and the angle of rotationθhave been foundcorrectly,but for
Fig.6d,the result is not very satisfactory.The reason for this is that the template Fig.3a cannotrepresent the face of interest at all times.That is to say,although the person to be recognized in different imagesis the same,the template cannot give the features for thisperson at all times(different appearance,etc.),and in allconditions.(The creation of the template is shown in Fig.4.)A second reason is that the algorithm itself has some problems.For example,by using a GA-based recognitionmethod,the settings of the search field(in this paper,(x,y,rate,?) is selected),the determination of the genetic
operations,and the selection and optimization of the fitness function all have a strong effect on the level of recognition of theresultant image.
Fig.4.Creation of template
For the purpose of comparing the effects of the GA-based algorithm,the result of the general matching method7is also presented.From Fig.5,we see that although both theoriginal image(the top-left image)and the template(thetop-right image)are simplified by binarization,the matching time is 1 min 22 s.The recognizable result is the bottomleft image in Fig.5.
Fig.5.Result of searching by a GA
4 Conclusions
In this article,the GA-based image recognition method istested,and a comparison with the general matching methodis presented.
As we know,the GA starts with an initial set of randomsolutions called
thepopulation.Each individual in thepopulation is called a chromosome,and represents a solution to the problem.By stochastic search techniques basedon the mechanism of natural selection and natural genetics,genetic operations(crossover and mutation)and evolutionaryoperations(selecting or rejecting)are used to search forthe best solution.8 In this article,the chromosomes generated by the GAcontain information about the image,and we use the genetic operators to obtain the best match between the originalimage and the template.The parameters are the coordinates(x,y)of the center of the object in the original image,the rate of scale,and the angle of rotationθ.
In fact,translation,scale,and rotation are the three maininvariant moments in the field of pattern recognition.9However,for face recognition,the facial features are difficult toextract,and are calculated by the general pattern recognition theory and method.10Even these three main invariant moments will not be invariant because the facial expressionis changed in different images.
Thus,recognition only gives the best matching resultwithin an upper predetermined threshold.Both the GA-based method and the general template matching methodare presented here,and the comparison with the traditionalpattern matching method shows that
the recognition is satisfactory,although under some conditions the result is notvery good(Fig.6d).
Based on the results of the experiments described here,future work
willemphasize(i)optimizing the fields of chromosomes,and(ii)improving the fitness function by addingsome terms to it.This work is important and necessary inorder to improve the GA-based face recognition system.
References
1.Sugisaka M,Fan X(2002)Development of a face recognitionsystem for the life robot.Proceedings of the 7th InternationalSymposium on Artificial Life andRobotics,Oita,Japan,vol 2,Shubundo Insatsu Co.Ltd.,pp 538–541
2.Castleman K(1998)Digital image processing.Original editionpublished by Prentice Hall;a Simon&Schuster Press of TsinghuaUniversity,China
3.Iba H(1994)Foundation of genetic algorithm:solution of mysticGA(in Japanese).Omu Press
4.Deguchi K,Takahashi I(1999)Image-based simultaneous controlof robot and target object motion by direct-image interpretation.Proceedings of the 1999 IEEE/RSJ International Conference onIntelligent Robot and Systems,Kyongju,Korea,vol 1,pp 375–380
5.Jaehne B(1995)Digital image processing:concepts,algorithms,and scientific applications,3rd edn.Springer Berlin,Heidelberg,Germany
6.Agui T,Nagao T(2000)Introduction to image processing usingprogramming language C(in Japanese).Shoko-do Press 7.Ishibashi’s studying room of C++(in
Japanese).http://homepage3.nifty.com/ishidate/vcpp.htm 8.Gen M,Cheng R(1997)Genetic algorithms and engineering design.Wiley-Interscience,New York
9.Agui T,Nagao T(1992)Image processing and recognition(inJapanese).Syokoudou Press 10.Takimoto H,Mitsukura T,Fukumi M,et al.(2002)A design of aface detection system based on the feature extraction method.Proceedings of the 12th Symposium on
相关推荐：
loading