• No results found

An Improved Face recognition Technique

N/A
N/A
Protected

Academic year: 2020

Share "An Improved Face recognition Technique"

Copied!
5
0
0

Loading.... (view fulltext now)

Full text

(1)

An Improved Face recognition Technique

Debjyoti Bagchi

Asst. Prof, Calcutta Institute of Engineering and Management E-mail: [email protected]

Debasish Chatterjee

Student , Calcutta Institute of Engineering and Management E-mail: [email protected]

Abstract

We are proposing an approach for identifying face from given database which stores large amount of distinct faces. Database is considered that stores three unique parameters for each face. Based on these, the database is divided into few categories and sub-categories .Whenever it is required to search a face, computation of the first category that contains the face is performed, based on first parameter attached with the face. In this way the next category can be found based on the other parameters .The final category may contain many faces and finally another search technique is performed to find out the actual face. It produces better result in terms of time consumption comparing to other popular approaches. We also use neural network technique to find out whether the picture to be searched is a face or not and also the concept of fuzzy logic to calculate the database categories.

Keyword

Face Recognition; Application of fuzzy logic; image processing. Introduction

(2)

1. Pupil to pupil distance 2. Pupil to nose-tip distance 3. Size of lips

4. Distance between two cheek bones, etc[1,8]

Methodology

Before we apply our algorithm, we need to store the image and few parameters associated with the image in a database as explained in figure.

Figure: 1

Where m is the address of the next range n is the address of the corresponding 2nd pass or database. The first column represents the ranges of the first pas .The second column represents the ranges of the second pass. The third column represents the database.

To manage the database we considering that there exist C groups or classes denoted by W1,W2…..Wc and associated with each pattern x is a categorical variable z that denotes the class or groups membership that is if z belongs to i then the image belongs to Wi where i belongs to (1,2….c) .

Figure: 2

(3)

we conceder the length of the picture in database is L and pupil to pupil distance is P and length of the detected face and pupil to pupil are L1and , P1from neural network then

1 1

P

P

L

L

;





1 1

*

L

L

P

P

. (1)

So after all the size adjustment we need to choose the parameters for comparison those are

1. Pupil to pupil distance.

2. Distance from the pupil to the nose tip. 3. Size of the lip.

These values are calculated these standard codes and we store these values with the face in the database. Then we feed this as input to the fuzzy logic based face recognition system which produces a value corresponding to the input image.

Set of Subset of

Pictures to pictures to

Input be searched be searched

Input Image + + Image

Input image Input image

Figure: 3

After the above procedure we use the following algorithm for face detection.

Algorithm

1. Calculate the pupil to pupil distance, the distance from the pupil to the nose tip and the size of the lip. 2. Check if the obtained pupil to pupil distance lies in the first sub-range. If yes, go to step 4. If no, go to

step 3.

3. Repeat step 2 to check if the pupil to pupil distance lies in the next sub-range.

4. Check if the obtained pupil to nose tip distance lies in the first sub-range. If yes, go to step 6. If no, go to step 5.

5. Repeat step 4 to check if the pupil to nose tip distance lies in the next sub-range.

6. Check if the obtained length of the lips lies in the first sub-range. If yes, go to step 8. If no, go to step 7. 7. Repeat step 6 to check if the length of the lips lies in the next sub-range.

8. Now we have obtained the pictures to be searched. So this means we have successfully ignored nearly 90% of the pictures through the above steps. Now we need to search the remaining pictures using the values calculated in step 1 as parameters. For this we use linear/hash searching algorithm.

Result and Conclusion

To derive the conclusion of this approach first we explain a mathematical approach for the above algorithm. In our technique we use divide and conquer algorithm and hence we calculate the time complexity for face detection algorithm. Generally in divide and conquer algorithm we divide the total database into small instances recursively up to level 3 and calculate the complexity for each instance and then combines the complexity to calculate the total complexity.

The following tree represented the total image database and its distribution. Fuzzy box 

with 

parameter 

pupil to pupil 

distance 

Fuzzy box 

with 

parameter 

pupil to nose 

tip distance 

Fuzzy box 

with 

parameter 

length of 

the lips 

Final set of 

picture to 

be 

(4)

Figure: 4

If the total image is p, then total approximate image in a group is (p/10). So if we choose a partition then total (9p/10) image can be ignored. If we apply this procedure for three levels the total image in a group are p/1000 (approx). We can ignore (999p/1000) images. Now if we apply linear search /binary search/hashing procedure with the final set then the searching time is less.

Total time complexity = O(p/1000) + x

x is the time complexity to determine the final set which contains the image. The resulting complexity of x can be defined as

T (n) =3 * T (n/10) + Cn

This process adds some linear complexity Cn. Then, T (n) = O(n log n).

The following graph represents the comparison between image recognition time from the database using binary search linear search and our proposed approach. This graph is also represents that both binary search and linear search technique takes more time to identifying an image compare to the above described algorithm.

No of Elements Figure: 5

If the number of image in the database is greater than 100000 then we divided the database into smaller sub set and the algorithm is applied to each subset recursively .This approach take some additional time but still it will take less time compare to other algorithms for searching a image from same number of data. As the graph

Reco

g

n

itio

n Ti

(5)

occur while implementing the algorithm, will not affect the result much. However further studies can be done to decrease the computational complexity involved with Fuzzy Logic.

References:

[1] Bonsor Kevin, Johnson Ryan. “How Facial Recognition Systems Work” http//electronics.howstuffworks.com/gadgets/high-tech-gadgets/facial-recognition1.htm

[2] Li Yi, Zhang Baochang, Shan Shiguang, Chen Xilin, Wen Gao. “Bagging Based Efficient Kernel Fisher Discriminant Analysis for Face Recognition”

[3] Paschalakis Stavros*, Bober, Miroslaw “Real-time face detection and tracking for mobile videoconferencing” , Mitsubishi Electric ITE B.V. Visual Information Laboratory, The Surrey Research Park, 20 Frederick Sanger Road, Guildford, Surrey GU2 7YD, UK [4] Qayyum Usman. “Efficiency Enhancement of Neural Network with Phase Only Correlation”. NESCOM, Islamabad, Pakistan.

[email protected]

[5] Romdhani Sami, Torr Philip, Bernhard Sch ¨olkopf3 and Andrew Blake4, “Efficient face detection by a cascaded support-vector machine expansion”,the royal socity 10.1098/rspa.2004.1333

[6] Sajid I, Ahmed M.M. and Taj I. “Time Efficient Face Recognition Using Stable Gram-Schmidt Orthonormalization” Internationa journal of signal processing and pattern vol 2 No.1 March 2009.

[7] Sajid, I.; Ziavras, S.G.; Ahmed, M.M.” Hardware-Based Speed Up of Face Recognition Towards Real-Time Performance” This paper appears in: Digital System Design: Architectures, Methods and Tools (DSD), 2010 13th Euromicro Conference on Issue Date: 1-3 Sept. 2010 On page(s): 761-3 - 770

Location: Lille Print ISBN: 978-1-4244-7839-2 INSPEC Accession Number: 11626567 Digital Object Identifier: 10.1109/DSD.2010.45 Date of Current Version: 01 November 2010

[8] Tanaka James a, *, Giles Michael a, Kremen Sarah b and Valerie Simonc “Mapping attractor fields in face space: the atypicality bias in

face recognition r”. Volume 68, Issue 3, September 1998, Pages 199-219

References

Related documents

Methods: The boundary conditions for field analysis by the internal Dirichlet problem are introduced, based on the vector potential field excited by external current coils..

Since the main dish is a symmetrical paraboloidal structure, the rotation of the sub-reflector about z -axis (see Figure 3) does not actually modify the radiation pattern, but

 Leslie  Bisson,  for  providing  the  vision  and  leadership  to   move  clinical  research  forward,  and  our  department  for  giving  us  the  resources  to  help  

This communication presents the results of an objective evaluation of HC3 version 4 (HC3v4), HC3 version 5 (HC3v5) and MACC-RAD against high quality measurements of fourteen

Chapters 6 to 8 describe the three principal studies that provide evidence to support the thesis statement; each of these studies considers a different facet of rational

Victor and Cullen (1988) recognized that the various ethical climate types are dependent upon what is perceived by the members as ethical. 2) The research questionnaire

Regions of 419 bp (Or47b), 736 bp (Or67d), 637 bp (Or65a), and 465 bp (Or88a) all drove reporter gene expression in patterns similar to those of the sensilla in which they

The paper discusses the satisfaction with services provided by South Africa Revenue Service (SARS) to public sector tax practitioners to enhance tax compliance.. A survey of 375