International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 5 A Novel Method for Rotational and Pose Invariant Child orAdult Age Classification based on Morphological Pattern Representation Schemes 1 Dr. Pullela S V V S R Kumar, 2 Ms. A V Lakshmi Dhanisetty, 3Dr, U Ravi Babu 1 Professor of CSE, Sri Aditya Engineering College, Surampalem, pullelark@yahoo.com *2 Computer Science & Engineering, Sri Aditya Engineering College, Email ID : srilakshmimca4@gmail.com 3, Professor, Dept. of CSE,MREC(A), India, uppu.ravibabu@mrec.ac.in Abstract-- The present paper proposes an approach for classify the facial images either child or adult based on local texture features extracted on facial images. The local texture features used in this paper are Morphological-prehistoric Patterns with grain components (MP-g). these patterns are extracted on a Local Directional Pattern (LDP) values. Edge response values in all eight directions from a 3×3 local window are used to calculate the LDP values. The local descriptor LDP is more constant in the incidence of noise and lighting changes, since edge response level is more stable than pixel intensity. The proposed method is rotationally& poses invariant when compared to pattern trends that represents a shape. The proposed method is tested on large set of images from different databases like FgNet, Morph, Google and scanned images. The present paper proves the efficiency of the proposed method. Indexed Term— Morphological Prehistoric Patterns, Local Directional Pattern, invariant, pose invariant. 1. INTRODUCTION Computer vision and psychophysics researchers faced so many challenges by face recognition of human and representation the typical features of human faces. Human faces be in the right place to a 3-D objects. Therefore, to develop an accurate representation for description that consider for lighting; face variations, facial expressions, etc.; of facial images. In face recognition system, large database maintained and search an image in that database. Day to day increase the size of the database and also increase the computational cost for the classification of human face into two categories i.e. either child or adult. To address this problem some procedures that highlights the important of facial development over a period of instance and considered a representation to minimize the dissimilarity between testdatabasesand images which are used at the time of method development. In appearance-based method, a face image is usually considered as a point in the high-dimensional space. Many linear subspace learning methods, such as LDA/FKT (linear discriminant analysis/Fukunaga–Koontz transform) [6], Eigenface [1, 2], C-LDA (complete LDA) [4], Fisherface [3], MMSD (multiple maximum scatter difference) [5], and Laplacian face [7] are typical dimensionality reduction methods to find a low-dimensional feature space. First,Local Binary Pattern LBP operator was introduced by Ojala et al. [8] for texture categorization and texture analysis. The LBP given good results especially in texture analysis so that LBP mostly suitable in texture examination and its applications. The LBP operator has hugeacceptance against lighting changes. The main advantage of LBP is, It was obtained with sample calculations. From the above properties, it is well suitable for real-world applications like image texture analysis. The concept of LBP was also extended in applications such as face recognition and age classification [9, 10,11]. In above proposed methodsareapplied on small set images and get comparative results but those methods are not suitable for all kinds of databases. The proposed method is applied on large set of images and different kinds of databases. No such method is available to classify the facial images into two categories i.e. either child or adult in effective and efficient manner of different database like FgNet, Morph Google and scanned image databases. The paper is organized as follows: Section 2 explains the proposed method and result analysis, experimental details and discussions are explained in Section 3. Conclusion part are given in the section 4. 2. T HE PROPOSED MORPHOLOGY P REHISTORIC P ATTERNS WITH G RAIN COMPONENTS ON LDP FOR CHILD AND ADULT AGE C LASSIFICATION The present paper identified MP-g based on changes in the facial skin. The skin of a human face tends to more alters with growing age is observed by the present method. These changes in the skin are subjugated by MP-g. The other important feature of the present method is to find which MP-g causes, the quick changes in the skin that leads to a higher age classification rate. The proposed novel scheme contains fivemajor steps. In the first step, the facial color image is cropped to cover only skin area of the face. Convert the RGB into gray scale image in 2nd step. Gray level facial image is converted into a two valued image by using LDP approach in the 3rdstep. In the fourthstep, identify the Morphological prehistoric patterns with one to eight-grain components (MP1g to MP-8g). In the final step, derive an algorithm for classifying the facial image into either child or adult. The schematic diagram of the entire process is illustrated in figure 1. 153805-7474-IJECS-IJENS © October 2015 IJENS IJENS International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 6 Cropping Original Facial image Cropped Image Binary facial Image Identify MP-1g to MP-8g patterns Convert RGB to Gray LDP with RobinCompass MAsk Classification Fig. 1. Block diagram of the child and adult classification system Step 1: Cropping: In this proposed method cropping is necessary because of identifying the facial skin edges. The proposed method based on the skin area of the human face. The given database, each facial image consist not only skin area but also hair, neck, and other areas also included. So to remove the unwanted area from the facial image cropping is necessary for effective results. The input facial image is cropped to cover the entire skin area of the face based on the location of two eyes in the first step as shown in figure 2. (a) (b) Fig. 2.Crop operation a) Input image b) Output image after cropping Step 2: RGB to Gray scale conversion: So many color models are available in color image processing. The facial image is converted to gray scale image for identifying the morphological patterns on the facial color image. The present paper uses HSV color model for converting the facial color image into gray scale, because the present study is aimed to classify the human age into four groups with a gap of 15 years based on changes on facial skin are identified on gray scale image.HSV color space separates the color into three categories i.e. hue, saturation, and value. Separation means variations of color are observed individually. The convertingequations for RGB to grey level conversion are given below. ( ) (1) ( ) S= (2) (3) (4) (5) In this work, the color component Hue (H) is considered as grey information for the classification of facial images. And this value is lies between 0 and255. Step 3: Local Directional Pattern Local Directional Pattern (LDP) [12] concept is used in this present study because it has more advantages compare toLBP approach. The LDP approach is more suitable for age group classification because this approach considers the edge response values in all different directions instead of surrounding neighboring pixel intensities like LBP. This provides more consistency in the presence of noise, and illumination changes since edge response magnitude is more stable than pixel intensity. The LDP is based on LBP. In the LBP operator, a gray-scale invariant texture prehistoric, has gained significant popularity for describing the texture [13]. It labels each pixel of an image by thresholding its Pneighboring values with the center value by converting the result into a binary number by using Equation 6. p 1 LBPP ,R ( xc , yc ) s( ESP CP)2 p , p 0 1 x 0 s ( x) 0 x 0 (6) Where CP denotes the gray value of the center pixel (xc, yc) and ESP corresponds to the gray values of P equally spaced pixels on the circumference of a circle with radius R. Local Directional Pattern with Robin Compass Masks Response The LDP generates eight-bit binary code assigned to each 3×3 sub window of an input image. These patterns are calculated by comparing the relative edge response value of a pixel in different directions by using Robin compass masks. The Robin compass masks in eight distinct orientations (r0~r7) centered on its own position. The Robin compass masks are shown in the Fig.3. 153805-7474-IJECS-IJENS © October 2015 IJENS IJENS International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 7 Fig. 3.Robin Compass masks in eight directions. Applying Robin masks on 3×3 masks, eight mask values V0, V1, …,V7 are obtained, each representing the edge significance in its respective direction. The mask values are not equally important in all directions. In robin mask values corner or edge pixels show higher values because those pixels are more important in particular direction. The LDP code produces the more firm pattern in the presence of noise, illumination changes and various (a) 26 38 85 10 50 53 48 32 60 LBP=00111000 LDP=00010011 conversion schemes of color facial images into gray images. For example, Fig. 4 shows an original image and the corresponding image with illumination changes. After illumination change, 5th bit of LBP changed from 1 to 0, Thus LBP pattern changed from uniform to a non-uniform code. Since gradients are more stable than gray value, LDP pattern provides the same-pattern value even in the presence of noise and non-monotonic illumination changes. 26 38 85 10 50 49 48 32 6 LBP=00101000 LDP=00010011 (b) Fig.4.Calculating the LDP and LBP values in two different situations (a) Original Image (b) Image with Noise Step 4: Evaluation of Morphological-prehistoric Patterns with Grain Components (MP-g) on LDP On the binary LDP facial texture images of the previous step, the present study evaluated the incidence of MP-g on a 3×3 mask. The present study classify the facial images into either Child or adult based on the number of grains incidence on a 3×3 sub-window LDP facial image in any orientation. Thismakes the present method as rotationally and poses invariant. The present method countsfrequency incidence of MP-g if and only if the central pixel of the 3×3 window is 1 and it is treated as a grain. If the central pixel is zero (0) then 3×3 window is treated as not-a-grain,In the following figures „0‟ indicates no grain and „1‟ indicates a grain. There can be eight combinations of MP-1g, which are shown in the Fig. 5. By any rotation, the MP-1g may change its position in 8 ways 0 0 0 0 1 0 (a) 0 0 1 1 0 0 (e) 0 1 0 0 0 0 0 0 0 0 1 0 (b) 1 0 0 1 0 0 0 0 1 0 0 0 on a 3×3 mask as shown in Fig.5. The present method counts the frequency incidence of MP-1g on a 3×3 mask irrespective of its position. Therefore, the present method is rotationally invariant. There will be seven different formations of MP‟s with 2pixel-grain components (MP-2g) by fixing one of the grains at pixel location (0,0) on a 33 mask as shown in Fig. 6. In a similar way, there will be six formations of MP-2g by locating one of the grains at the pixel location (0,1) as shown in Fig. 7. Thus, there will be 7! ways of forming MP-2g for a 3x3 window. In the same way, there will be 6!, 5!, 4!, 3!, 2! and 1! ways of forming MP-g of 3, 4, 5, 6, 7 and 8 respectively, on a 3x3 mask irrespective of their rotational positions. 0 0 0 0 1 1 ( c) 0 1 0 1 0 0 0 0 0 0 0 0 (f) (g) Fig. 5. Representation of 3×3 window withMP-1g. 153805-7474-IJECS-IJENS © October 2015 IJENS 0 0 1 0 1 0 (d) 0 0 0 1 0 0 0 0 0 1 0 0 (h) IJENS International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 1 0 0 0 1 0 0 1 0 1 0 0 (a) 0 1 0 0 0 1 1 0 0 (b) 1 1 0 0 1 0 0 0 0 0 1 1 0 0 0 1 0 1 ( c) 1 0 0 (e) 1 1 0 0 0 0 0 1 0 8 0 0 0 (d) 1 0 0 (f) 0 1 0 1 0 0 (g) Fig. 6. Representation of MP-2g by fixing grain at (0,1) 0 0 0 1 1 0 (a) 0 1 0 1 1 0 0 1 0 0 0 0 0 0 0 1 1 0 (b) 0 1 1 1 0 0 0 0 1 0 0 0 0 0 0 1 1 1 ( c) 0 1 0 1 0 0 0 0 0 1 0 0 (d) (e) (f) Fig. 7. Representation of MP-2g by fixing one of the grain component at (0,1). The frequency incidence of MP-1g to MP-8g are computed by the proposed Algorithm 1 on a 33 nonoverlapped mask of LDP facial texture images, and stored in the facial database. The pseudo code entire method is illustrated below Step 1: Read input image Step 2: Crop the facial image Step 3: if the input image is RGB then covert the image into gray level image as described by the step2 in section2 with the size of N×N Step 4: Define the RobinMasks(r0, r1, r2, r3, r4, r5, r6, r7 ) based on masks given in the figure 3. Step 5: Convert the Gray image into Binary by using LDP Robin Masks (a) Apply 2-D convolution of image G(i,j) and mask MN(i,j) with size 3×3 F(i,j)= G(i,j)*Mn(i,j) (b) Calculate the threshold, th= F(i+1,j+1) (c) Initialize the variables p=0 and q=0 (d) ifF(p,q)>=thenBW(p,q)= 1 otherwise BW(p,q)= 0 (e) Initialize the variables l=0 and m=0 (f) for a = l to l+2 and b = m to m+2 do the following procedure (i) if BW(l+1,m+1)==1 then Count the frequencies of incidence of Grain Features from 1,2,…,8 otherwise update m = m+3 (ii) ifm< N then go to step (f) otherwise update l = l+3 and m = 0 (iii) if l< N then go to step (f) otherwise Store the Grain features in the face recognition database and stop the procedure 3. RESULTS AND D ISCUSSION To find the consequence of the present method, the present study has evaluated MP-1g to MP-8g on facial LDP images of child and adults from different poses of 1002 FgNet ageing databaseimages, 12000Morph data base images, 1500 scanned images and 750 Google imagestotally, its leads of 15252 images. Some of the child and adult images in different databases are shown in figure8 and Figure9. The present study considered that the childhood agesbetween 0 and 15 years and adulthood is over30 years. Table 1 and Table 2 indicate the frequency incidence of MP-g for sample database of FgNet. The facial image is recognized as child or adult based on Algorithm 1. From the computed frequency of incidence of MP-1g to MP-8g by the Algorithm 1 the present study observed that only twoMP-gS are exhibiting successful child-adulthood classification rates. The MP-1g and MP-4g have proved to have significant, precise and accurate classification rates than others MP-g‟s. The present study suggests that it is not necessary to evaluate frequency incidence of MP-2g, MP-3g, MP-5g, MP-6g, MP7g and MP-8g on facial LDP images of the child and adult for classification. Since the facial images are of different poses, the proposed method is posed invariant. To prove the proposed method is rotationally invariant the MP-g are evaluated with different rotations 30, 45 and 135 degrees 153805-7474-IJECS-IJENS © October 2015 IJENS IJENS International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 andthe obtained results are scheduled in Tables from 3 to 8. Even by rotation with different angles, the Algorithm 1 based on MP-1g and MP-4g classifies the child and adult. This proves that the present method is rotationally invariant. Thus, 9 the present method has overcome the disadvantage of pattern based and also previous methods that are rotational and pose variant. Fig. 8. Sample facial images of adults with different poses of FgNet aging database. Table I Frequency incidence of child facial images using MP-g. Child images 015A01 002A03 009A03 069A03 053A06 066A06a 019A07 016A08 023A09 073A09 065A09 011A11 022A11 012A12 008A13 Sci-img-01 Sci-img-02 Sci-img-03 Sci-img-04 Sci-img-05 mor-img-01 mor-img-02 mor-img-03 mor-img-04 mor-img-05 MP-1 14689 15365 13649 13451 14478 19661 14369 17206 21243 10871 11821 18020 22196 16829 14843 13568 14768 15649 18637 19037 17433 15684 13081 16566 14595 MP-2 765 864 752 1468 1090 1170 1092 788 1459 1186 806 958 789 923 1421 963 768 854 913 963 876 961 1293 620 926 MP-3 679 837 913 1188 861 929 982 658 990 1164 661 687 799 690 1046 914 863 786 597 618 723 831 1176 640 726 MP-4 964 952 994 1228 1052 1138 1163 1009 1183 1122 856 919 910 1023 1162 984 913 927 1013 1019 1101 978 1164 917 935 MP-5 1047 964 899 2209 1233 1454 1676 833 1129 2258 930 733 1092 854 1305 865 1025 973 918 879 1123 1079 2372 1065 1153 153805-7474-IJECS-IJENS © October 2015 IJENS MP-6 2015 2135 2049 4066 1763 2221 3527 1216 1521 4396 1774 1153 1325 9080 1389 1423 1564 1579 1679 2123 2094 1998 4605 2490 2255 MP-7 1371 1129 1078 2155 1186 1379 2956 363 774 3901 1501 611 863 617 746 865 1365 1278 1135 1496 1526 1428 2996 2168 2626 MP-8 2076 3168 7045 895 548 828 3015 138 581 2482 10531 230 706 195 299 203 276 536 895 9563 8768 1963 1493 4414 5664 IJENS International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 10 Table II Frequency incidence of adult facial images using MP-g. Adult images 027A41 062A41 071A42 033A44 047A45 028A46 045A48 003A49 048A50 039A52 004A53 005A61 006A61 Sci-img-101 Sci-img-102 Sci-img-103 Sci-img-104 Sci-img-105 mor-img-501 mor-img-502 mor-img-503 mor-img-504 mor-img-505 MP-1 13658 15638 17698 18486 22360 20731 18649 20134 23498 16538 18649 16498 15253 22308 20125 21237 19858 14628 19556 14390 19679 20860 23198 MP-2 1075 1349 1745 2146 1140 861 1948 1956 896 2313 2013 2146 1594 1374 2237 2379 1290 1643 1065 2159 2140 1351 1298 MP-3 865 913 1035 1583 924 760 964 813 854 2169 979 1237 1477 981 1165 1179 1067 1629 910 1912 1204 952 1076 MP-4 1364 1268 1495 2037 1273 1182 2018 2134 1204 2406 1769 1649 2057 1437 1548 1788 1621 1936 1278 2513 1499 1771 1254 MP-5 1123 1235 1465 2140 1088 1368 2013 2348 1065 2635 1645 2156 2424 1105 1652 1847 1569 2414 1395 3067 1769 1637 1286 MP-6 1450 3623 2359 2855 1400 2120 1365 1433 1637 3013 1956 1765 3758 1233 2016 2347 2188 4082 2373 4187 2246 2585 1654 MP-7 1211 1466 1355 923 514 1011 1217 1175 885 1143 1259 1567 1534 251 1468 1346 852 2934 1587 1770 1646 998 586 MP-8 384 295 406 406 281 647 594 712 337 359 535 478 783 191 776 649 435 1310 716 578 662 522 224 Algorithm 1: Rotational and pose invariant child and adult age classification using MP-g on LDP. if (MP-1g<=13700) print (facial image is of Child) elseif (((MP-1g >14000) && (MP-1g <22300)) && ( MP-4g<=1200)) print (facial image is of Child) elseif (((MP-1g >14000) && (MP-1g <23500)) && ((MP-4g >1300) && (MP-4g < 2350))) print (facial image is of the Adult) else print (facial image is not of the child and the adult) end The Algorithm 1 classifies child from adult, based only on the 14000 to 23500 and a MP-4g count is less than 1300 then it frequency incidence of MP-1g and MP-4g values. If a MP-1g classifies as a child otherwise adult as specified in the value is less than 13700 then the facial image is treated as Algorithm 1. The same thing is also true for all rotational child, else they form group 2 entries. By considering both facial texture images. MP-1g and MP-4g values if a MP-1g count is inbetween 153805-7474-IJECS-IJENS © October 2015 IJENS IJENS International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 11 Table III Frequency incidence of a MP-g with 300 rotation on LDP for child facial images. Child images 002A03 008A13 009A03 011A11 012A12 015A01 016A08 019A07 022A11 023A09 073A09 069A03 066A06a 065A09 053A06 Sci-img-01 Sci-img-02 Sci-img-03 Sci-img-04 Sci-img-05 mor-img-01 mor-img-02 mor-img-03 mor-img-04 mor-img-05 MP-1 10346 9768 10769 15722 16014 7568 8316 9769 10325 12356 6106 14389 6139 7024 11737 13645 16819 13971 11261 11455 8301 14290 9658 8923 13257 MP-2 613 718 834 794 1268 513 648 716 839 746 1029 911 685 539 480 943 852 1431 984 585 1149 950 579 732 1595 MP-3 634 728 531 562 888 1365 1230 1065 943 1032 1094 785 734 517 471 978 662 1085 788 876 1227 751 528 567 1287 MP-4 1064 978 867 637 938 846 813 679 681 1035 1138 932 1010 617 743 1094 842 1169 999 1113 1004 992 663 703 1145 MP-5 1348 1346 927 774 889 1035 778 684 1235 1365 2230 1199 1349 756 982 1649 994 1196 1179 1869 2370 1268 673 765 2333 MP-6 2216 2575 1567 799 914 936 1365 1364 2346 2867 2255 1224 1374 781 1007 3121 1019 1221 1204 2348 2395 1293 698 790 2358 MP-7 2349 2659 1946 520 574 694 1346 1864 1964 2034 3884 2624 2439 1371 2062 2264 527 717 1038 2019 2722 1123 364 295 2002 MP-8 16494 13462 12346 18924 20005 19864 23156 23458 22467 19467 22269 31622 17932 30261 24097 18631 20343 19158 17023 17345 21617 20056 12336 11841 25244 MP-7 1425 1268 1361 670 580 1345 1227 1278 3038 1994 295 419 1070 1236 943 1600 957 1649 1721 1833 1629 1267 603 MP-8 23456 22658 31456 32992 30986 20349 27725 25648 31314 21546 24098 24834 27388 20658 33341 33123 35323 26690 27988 26485 24845 26921 26817 Table IV Frequency incidenceof aMP-g with 300 rotation on LDP for adult facial images. Adult images 003A49 005A61 006A61 027A41 028A46 033A44 039A52 045A48 047A45 048A50 071A42 062A41 004A53 Sci-img-101 Sci-img-102 Sci-img-103 Sci-img-104 Sci-img-105 mor-img-501 mor-img-502 mor-img-503 mor-img-504 mor-img-505 MP-1 11235 13465 12346 17960 18893 13495 15144 10395 10970 12586 18556 18032 14167 13644 15264 9034 11630 10349 12453 11358 11173 14159 14129 MP-2 1358 1649 1346 812 1196 574 2391 974 1526 618 1167 1081 833 529 1137 2093 1679 936 817 928 1693 916 1234 MP-3 1265 1349 1765 1326 1280 1270 2726 1794 3704 1873 910 1110 1972 1979 1905 3372 1973 1864 1945 2034 3195 1957 1716 MP-4 1547 1463 1429 1823 1399 2214 2318 1358 1850 2199 1296 1246 1967 2035 1213 2285 1607 1647 1950 1713 1942 1999 1339 MP-5 1649 1563 1429 1034 1255 1099 2624 1037 2327 1267 1155 1008 1274 1348 1471 2930 1783 945 865 765 2371 1343 1530 153805-7474-IJECS-IJENS © October 2015 IJENS MP-6 2065 2164 2318 652 980 1649 2114 2495 1540 1765 888 835 694 1649 995 1832 1317 2020 1367 1428 1517 803 997 IJENS International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 12 Table V Frequency incidence of a MP-g with 450 rotation on LDP for child facial images. Child images 002A03 008A13 009A03 011A11 012A12 015A01 016A08 019A07 022A11 023A09 073A09 069A03 066A06a 065A09 053A06 Sci-img-01 Sci-img-02 Sci-img-03 Sci-img-04 Sci-img-05 mor-img-01 mor-img-02 mor-img-03 mor-img-04 mor-img-05 MP-1 12649 7510 13432 9468 12466 15160 9846 11235 8697 10356 5228 8764 13568 14665 12356 13091 8091 7264 6497 14763 15992 10398 5298 6109 10904 MP-2 815 1243 946 798 1638 1270 575 1035 661 897 1067 596 929 1123 946 1475 714 613 678 784 865 1030 660 592 485 MP-3 1344 1294 826 1037 1397 873 613 1175 535 1239 1183 915 837 1094 1211 1121 630 846 769 609 691 829 740 554 526 MP-4 611 1035 1065 864 1183 914 865 599 579 798 1194 1403 981 764 849 1181 735 943 1097 665 813 1013 1035 618 757 MP-5 1214 2210 1115 1679 2153 864 894 2133 601 2041 2169 1035 1176 1449 1365 1126 642 1863 1486 742 933 1062 1347 736 901 MP-6 1458 2766 1169 1864 2008 731 1649 2213 274 1946 3836 1764 2537 2103 2317 881 403 2031 1946 451 673 1039 2394 1344 2153 MP-7 1127 2771 1174 1095 2013 736 1465 845 279 799 3841 1564 2542 911 1094 886 408 1346 1294 456 678 1044 2399 1349 2158 MP-8 18188 22845 21530 18465 27594 21532 24683 20316 13086 17645 24038 23446 34117 19465 23157 20495 12142 26491 19485 20094 21366 17735 18580 32259 25795 MP-7 812 945 849 723 1477 2918 846 553 698 716 714 1652 1268 613 424 765 271 697 1034 465 582 1418 774 MP-8 15546 19468 23095 16866 23837 29048 23456 24003 16489 23524 23456 27397 23110 21354 26558 20164 27478 18408 18465 17563 19738 31735 23089 Table VI Frequency incidence of a MP-g with 450 rotation on LDP for adult facial images. Adult images 003A49 005A61 006A61 027A41 028A46 033A44 039A52 045A48 047A45 048A50 071A42 062A41 004A53 Sci-img-101 Sci-img-102 Sci-img-103 Sci-img-104 Sci-img-105 mor-img-501 mor-img-502 mor-img-503 mor-img-504 mor-img-505 MP-1 10349 16542 15612 12654 9332 11751 14687 17491 13456 18249 12145 12295 14627 10346 19090 13464 19684 12062 13459 14654 11687 16286 14703 MP-2 864 1165 914 1326 1765 1283 1037 840 1034 483 1348 1660 775 584 1059 1256 1334 581 1360 867 1359 2554 1186 MP-3 649 1021 794 867 1637 1415 913 703 918 490 765 1637 741 596 908 864 1059 571 972 877 1068 2217 1012 MP-4 946 2156 1949 876 2043 1714 2300 1811 785 1624 648 1913 1921 663 1315 1550 1212 1672 1948 975 1356 2278 1296 MP-5 563 2564 1032 946 2516 2157 2013 776 1945 690 1762 2284 1112 703 952 1648 1080 783 1246 1645 1182 2400 1256 153805-7474-IJECS-IJENS © October 2015 IJENS MP-6 1786 2035 1739 2349 3233 3618 1649 1094 3015 1064 1034 3218 2005 1325 1138 1766 958 1562 2649 2789 1652 2728 1668 IJENS International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 13 Table VII Frequency incidence of aMP-g with 1350 rotation on LDP for child facial images. Child images 002A03 008A13 009A03 011A11 012A12 015A01 016A08 019A07 022A11 023A09 073A09 069A03 066A06a 065A09 053A06 Sci-img-01 Sci-img-02 Sci-img-03 Sci-img-04 Sci-img-05 mor-img-01 mor-img-02 mor-img-03 mor-img-04 mor-img-05 MP-1 16908 14038 11282 9605 9022 13230 14458 6239 15688 16120 6154 8419 14332 7009 11791 10395 14656 12349 11256 10356 9468 8864 9862 10354 11349 MP-2 857 1642 1149 665 757 1650 1012 744 1292 1098 1259 1259 1020 639 554 1037 1123 978 981 1035 1210 971 968 811 722 MP-3 758 1104 856 550 651 1475 832 744 894 1201 1315 1315 787 570 529 594 913 795 981 596 637 779 891 730 616 MP-4 940 1144 956 613 693 700 948 1014 873 1175 1058 1058 1017 613 721 894 931 1027 1346 1294 1246 1495 1221 1323 1424 MP-5 889 1129 1061 582 644 2120 1177 1435 900 2164 2254 2254 1146 723 917 1013 1216 1324 1525 1646 2112 2001 1940 899 905 MP-6 1108 1426 1651 962 788 3777 2483 2865 1173 4652 4164 4164 2058 1543 2219 966 2034 2246 3156 4023 3947 3129 798 1039 1164 MP-7 574 595 1056 327 369 2491 2532 2362 652 3839 2819 2819 1140 1367 2208 1345 1649 1864 1447 1552 1336 1649 975 869 713 MP-8 22910 22186 19238 14585 13645 29253 35607 19941 23040 25542 24337 24337 23021 33761 27286 23456 22151 19457 16485 14863 13498 32467 23622 19468 17644 MP-7 775 599 1038 1465 626 1094 1036 669 504 1013 913 1711 1311 2917 745 844 240 966 399 1132 600 948 1414 MP-8 22156 18462 23141 19468 22165 20946 30865 20134 22529 17023 27895 26079 23498 27607 22240 23145 26201 26499 25279 21346 22196 22209 22547 Table VIII Frequency incidence of a MP-g with 1350 rotation on LDP for adult facial images. Adult images 003A49 005A61 006A61 027A41 028A46 033A44 039A52 045A48 047A45 048A50 071A42 062A41 004A53 Sci-img-101 Sci-img-102 Sci-img-103 Sci-img-104 Sci-img-105 mor-img-501 mor-img-502 mor-img-503 mor-img-504 mor-img-505 MP-1 13954 10580 9467 8879 14446 9846 15052 1465 16654 10902 12649 11128 10317 10686 17147 13495 18619 10346 17969 8649 13542 13500 8233 MP-2 1178 1353 1532 711 942 917 2492 1265 886 635 937 1824 943 1405 552 1399 1361 1434 1154 1034 1273 808 1795 MP-3 849 1035 741 951 812 756 2354 1094 692 551 963 1655 753 1425 488 794 1052 852 873 355 1012 711 1686 3.1 Comparison of Results with othersProposedmethods results: The efficiency of the present method is compared with others proposed methods like“geometric properties” approach [14] and SPBPLME[15] methods. Geometric properties approach MP-4 846 1273 943 694 1933 862 2335 946 1839 1649 864 1946 953 1771 1639 1067 1282 824 1227 761 1317 1991 2114 MP-5 916 1124 761 696 982 851 2356 813 759 763 945 2186 762 2017 633 1034 990 848 935 515 1223 1031 2305 MP-6 2167 1672 2011 1036 1752 853 2800 2034 1082 1474 1943 3201 1864 3750 1070 1846 1005 1765 1182 1946 1595 2035 3420 extracts the geometric facture of Wavelet facial image. The overall percentage of this method is 91.88%.The proposed SPBPLMEmethod estimated incidence rates for zero transitions on prominent binary patterns on SBPLME. The overall proportion of classification rate of the SPBPLME is 153805-7474-IJECS-IJENS © October 2015 IJENS IJENS International Journal of Electrical & Computer Sciences IJECS-IJENS Vol:15 No:05 about 96.13. The proportion of classification in each group of the proposed method and others methods are listed out in Table 9. The graphical representation of the classification results is shown in figure9. The Table 9 clearly indicates that the proposed method yields better performance rate when compared with the existing methods. Table IX Overall % of classification rates of the proposed method and others methods. Image Dataset FgNet Morph Scanned Google 100.00% 98.00% 96.00% 94.00% 92.00% 90.00% 88.00% 86.00% 84.00% Geometric properties Approach 91.34% 90.79% 89.15% 91.28% SPBPLME method Proposed Method 96.37% 95.46% 96.13% 96.67% 97.73% 96.96% 98.15% 97.97% Geometric properties Approach [6] [7] [8] [9] [ 10 ] [ 11 ] [ 12 ] [ 13 ] SPBPLME method [ 14 ] Proposed Method [ 15 ] 14 Zhang S. and Sim T., “Discriminant subspace analysis: a Fukunaga–Koontz approach,” IEEE Trans. Pattern Anal. Mach. Intell. 29 (10) (2007) 1732–1745 He X., Yan S., Hu Y., Niyogi P. and Zhang H, “Face recognition using Laplacianfaces,” IEEE Trans. Pattern Anal. Mach. Intell. 27 (3) (2005) 328–340. T Ojala, M PietikaÈinen, and D Harwood published a paper entitled “A Comparative Study of Texture Measures with Classification Based on Feature Distributions” in the Journal of Pattern Recognition, Year: 1996 volume: no : 29 Pages 51 to 59. M Chandra Mohan, V Vijaya Kumar and B Sujatha published a paper entitled "Classification of child and adult based on geometric features of face using linear wavelets," in the Journal of Signal and Image Processing, volume no: 1 and issue no:3, yeAr : 2010 Pages:211 to 220. "Texton Based Shape Features on Local Binary Pattern for Age Classification", Chandra Sekhar Reddy et al.; ,IJIGSP 2012, 7, 5460 Novel method of adult age classification using linear wavelet transforms", Chandra Mohan, VijayaKumar V., Venkata Krishna V., IJCSNS, pp:1-8, 2010. Abdel-Mottaleb M. and Elgammal A., “Face Detection in Complex Environments from Color Images,” IEEE International Conference Image Processing, pp. 622-626, 1999. Pantic M. and RothkrantzL.J.M.“Automatic analysis of facial expressions: the state of the art,” IEEE Trans. Pattern Analysis and Machine Intelligence, 22(12) (2000) 1424-1445 "Classification of child and adult based on geometric features of face using linear wavelets", Chandra Mohan M., Vijaya Kumar V., Sujatha B., IJSIP, vol.1, Iss.3, pp:211-220, 2010. “Novel Approach for Child and Adulthood Classification based on Significant Prominent Binary Patterns of Local Maximum Edge (SPBPLME)”Rajendra Babu .Ch, DrSreenivasa Reddy. E and DrPrabhakara Rao. B, I.J. Information Technology and Computer Science, 2015, 06, 30-37 Fig. 9. Comparison graph of proposed method with others methods 4. CONCLUSIONS The present paperproposed a novel method for age classification of child and adults based on the MP-gSon LDPfacial skin. The novelty of the present method is, it is rotationally, pose, noise, illumination invariant due to basic principles of LDP and the proposed MP-g. The present approach outlines that one need not necessarily evaluate the frequency of incidence of MP-2g, MP-3g and MP-5g to MP8g for the age classification. The MP-1g and MP-4g contains more textural and topological information of the facial skin, that is the reason these two texture features are classifying the child and adult. REFERENCE [1] [2] [3] [4] [5] Turk M. and Pentland A. “Eigenfaces for recognition,” Int. J. Cognitive Neurosci. 3(1) (1991) 71–86. Turk M. and PentlandA.“Pattern Recognition Learning and Thought,” pp.486, Pretice Hall, New Jersey, 1973. Belhumer P. N., Hespanha J.P. and Kriegman D.J. ” “Eigenfaces vs. Fisherfaces: recognition using class specific linear projection,” IEEE Trans. Pattern Anal. Mach. Intell. 19 (7) (1997) 711–720. Yang J. and Yang J.Y. “Why can LDA be performed in PCA transformed space,” Pattern Recognition 36 (2) (2003) 563–566. Song F., Zhang D., Mei D. and Guo Z., “A multiple maximum scatter difference discriminant criterion for facial feature extraction,” IEEE Trans. Syst. Man Cybern. Part B Cybern. 37 (6) (2007) 1599–1606. 153805-7474-IJECS-IJENS © October 2015 IJENS IJENS