30 views

Uploaded by munadil98

In this paper an optical filter is proposed to minimize error while detecting reference object from group of targets having nearness pattern.

save

You are on page 1of 4

TECHNIQUE FOR CLASS ASSOCIATIVE TARGET DETECTION

M. Fahad*, M. Hasan*, T. Husna*, Md. Rafiqul Haider* and Mohammed Nazrul Islam*

*

Department of Electrical and Electronic Engineering, Bangladesh University of Engineering and

Technology, Dhaka-1000, Bangladesh, E-mail: nazrul@eee.buet.ac.bd

A new weighted reference based class associative product. Also the discrimination between target and

joint transform correlation technique has been non-target object is not so high.

proposed for multiple target detection process.

Different channels are used for each member of the In this paper, a simplified class-associative technique

class to get the correlation output and then the is proposed where weighted versions of different

different channel outputs are summed up to get the member images of a class are processed in different

final correlation output. The phase encoding and channels using the SPFJTC technique. Here spatial

phase-shifting principle is used to eliminate the false plane weighting is used instead of Fourier plane

alarms and other extraneous signals. Finally, a new weighting which eliminates the problem of perfect

post processing technique is employed to enhance alignment of filter in the Fourier domain. In earlier

the discrimination ratio between target and non- methods, every time there is a change in the member

target objects. Simulation results are presented to of a class object, it requires a corresponding change

verify the efficiency of the proposed method. in the filter fabrication and weight value in Fourier

domain, and also perfect position of the filter in

Fourier plane. In case of spatial plane weighting this

1. INFORMATION problem can be easily eliminated and real time

operation can thus be achieved. Then all the

Optical pattern recognition may be broadly correlation outputs of different channels are

classified into two categories; namely, VanderLugt combined in a single output plane to obtain the final

type filter-based correlation [1] and joint transform output. Here the position of reference image in

correlation (JTC) [2]. A VanderLugt type correlator different joint images is kept fixed to avoid

requires a priori fabrication of the filter used in the overlapping of desired autocorrelation peaks of

correlation process, whereas the JTC technique can different channels. Finally by varying the weighting

be operated at the video frame rates [3]. However, a factor of different member images, almost equal

classical JTC suffers from poor discrimination, correlation peaks can be obtained in the output plane

strong zero-order peak and wide correlation width.

Among the various modifications of JTC technique,

the shifted phase-encoded fringe adjusted JTC 2. ANALYSIS

(SPFJTC) is found to be the superior one in terms of

producing single sharp delta-like correlation peak In a classical JTC, the reference image and input

for each target object with high discrimination scene image are placed side-by-side by using an

capability [3]. SLM to form a joint image. For class-associative

target detection, weighted reference images of

For classification of objects from a given class, different members of the class are placed in different

different filter-based techniques have been proposed joint images to process them in parallels in the

utilizing phase information from the Fourier respective channels. Let r1(x,y) and r2(x,y) represent

transform of each object in that class [4]. However, the two member images in a class, respectively, and

most of the existing class-associative JTC t1(x,y−y1), t2(x,y−y2) …... tn(x,y−yn) represent

techniques produce a pair of correlation peaks for different target images in the input scene. Then the

each target object and utilize at most 50% of the two joint image formed in two channels can be

input and output plane spatial light modulator expressed as

n

f11 ( x, y ) = α r1 ( x, y + y0 ) ⊗ φ ( x, y )

f1 ( x, y ) = α r1 ( x, y − y0 ) + ∑ ti ( x − xi , y − yi ) (1)

n (10)

+ ∑ ti ( x − xi , y − yi )

i =1

n

f2 ( x, y ) = β r2 ( x, y − y0 ) + ∑ti ( x − xi , y − yi ) (2) i =1

i =1 f12 ( x, y ) = −α r1 ( x, y + y0 ) ⊗ φ ( x, y )

where α+β=1. The value of α and β can be varied n (11)

depending on the energy content of the power + ∑ ti ( x − xi , y − yi )

spectra of the reference images to achieve equal i =1

correlation peaks. The corresponding joint power The JPS of these joint images are then subtracted

spectrum (JPS) of first and second joint images from one another and the resultant JPS is again

|F1(u,v)|2 and |F2(u,v)|2, respectively, are given by multiplied by the same phase mask and the FAF

n filter.

F1 (u , v ) = α 2 R1 (u , v ) + ∑ Ti (u , v )

[ ]

2 2 2

P1 (u , v ) = F11 (u , v ) − F12 (u , v ) Φ (u , v )

2 2

i =1

n Ti (u , v ) R1 (u , v ) Ti (u , v ) R1 (u , v ) (12)

+ 2α ∑ n

i =1 × cos[Φ ti (u , v ) − (uxi + vyi ) − Φ r1 (u , v ) − vy0 ] = 4α ∑ Φ ti (u , v ) − (uxi + vyi ) − ψ (u , v )

i =1 × cos Φ(u , v )

Ti (u , v ) Tk (u , v ) − Φ r1 (u , v ) − vy0

n n

+ ∑∑

× cos[Φ ti (u , v ) − (uxi + vyi ) − Φ tk (u , v ) − (uxk − vyk )]

PF1 (u , v ) = FAF1 (u , v ) × P1 (u , v )

i =1 k =1

k ≠i

(3) B(u , v )

n =

F2 (u , v ) = α R2 (u , v ) + ∑ Ti (u , v ) A(u , v ) + R1 (u , v )

2 2 2 2 2

Ti (u , v ) R1 (u , v )

i =1

n Ti (u , v ) R2 (u , v )

− 2α ∑ exp[ j (Φ ti (u , v ) − (uxi + vyi ) − Φ r1 (u , v ) − vy0 )]

× cos[Φ ti (u , v ) − (uxi + vyi ) − Φ r 2 (u , v ) − vy0 ]

n

i =1 × 2α ∑

i =1 × Φ ti (u , v ) − (uxi + vyi )

n n Ti (u , v ) Tk (u , v ) + exp − j × Φ 2

(u , v )

+ ∑∑ − Φ (u , v ) − vy

i =1 k =1 × cos[Φ ti (u , v ) − (uxi + vyi ) − Φ tk (u , v ) − (uxk + vyk )] r 1 0

k ≠i

(13)

(4) In this paper, B(u,v)=1 and A(u,v)≈10-4 are chosen to

In equations (3) and (4), the third term gives the eliminate the pole problem and to reduce the effect

desired correlation peaks, while others produce of noise component in the correlation plane.

unnecessary peaks and false alarms. Here the shifted Similarly the fringe adjusted power spectrum for the

phase encoded fringe adjusted JTC (SPFJTC) is second weighted reference image is given by

employed to get single peak per target and to

PF2 (u , v ) = FAF2 (u , v ) × P2 (u , v )

eliminate extraneous signals. For this purpose, a

phase mask Φ(u,v), of random phase distribution B(u , v )

=

A(u , v ) + R2 (u , v )

2

Ψ(u,v) from −π to π, is used in the Fourier plane.

Φ(u , v) = exp[ jΨ (u , v)] (5) Ti (u , v ) R2 (u , v )

Fourier transformation of the reference image and its n exp[ j (Φ ti (u , v ) − (uxi + vyi ) − Φ r 2 (u , v ) − vy0 )]

1800 phase shifted version are multiplied by the × 2β ∑

i =1 × Φ ti (u , v ) − (uxi + vyi )

phase mask and then inverse Fourier transformation + exp − j − Φ (u , v ) − vy × Φ (u , v )

2

images. (14)

S11 =| α R1(u, v) | exp[ jφr1(u, v) + jvy0 ]×Φ(u, v) (6) Here for class associative target detection, each

reference image of the class is correlated with the

s11 = α r1 ( x, y + y0 ) ⊗ φ ( x, y ) (7) input scene or the target images to get two

S12 =| α R1 (u, v) | exp[ jφr1 (u, v) + jvy0 ] correlation plane outputs. Finally the correlation

(8) plane outputs are summed up to obtain the final

× exp[ jπ ] × Φ (u , v) output.

s12 = −α r1 ( x, y + y0 ) ⊗ φ ( x, y ) (9)

where ⊗ denotes the convolution operation. These Finally a post processing technique is incorporated

two phase-encoded reference images are then to equalize the target peaks and minimize the non-

combined with the input scene image to form two target peaks and noise. The correlation output is

joint images as given by

427

multiplied by the a filter whose characteristic technique performs well and the proposed filter

equation is given by further improve its performance.

correlation output plane and the parameter ‘shift’ is

determined by trial and error basis. The ‘shift’

parameter is used to make the target correlation

peaks almost equal whose magnitudes are greater

than the ‘shift’ parameter and diminish the non-

target peaks whose values are less than the

parameter ‘shift’. Thus the filter ensures the better

discrimination in the output plane.

Figure 1: Joint image of the first channel.

3. SIMULATION RESULTS

Performance of the proposed technique is

investigated by computer simulation using binary

images of English characters. The letters ‘E’ and ‘M’

are used as the selected targets and members of a

class. In the joint images shown in Fig. 1 and Fig. 2,

the characters ‘E’ and ‘M’ placed on the left side

represent the reference images of the first and second

channels, respectively, and the characters ‘E’, ‘T’,

‘H’ and ‘M’ on the right side of both channels

Figure 2: Joint image of the second channel.

represent the input scene images. Figures 3 and 4

show the correlation outputs of the first and second

channels, respectively, using the FAF technique. The

weights used in the first and second channels with

the reference images of the class are α and β,

respectively, to get equal correlation peak in the

combined correlation output plane, where α=0.52

and β=1−α. Fig. 5 shows the combined correlation

output after linear addition of the correlation outputs

of the two channels. From Fig.s 3, 4 and 5 it is

evident that due to the addition of correlation outputs

of two channels, the non-target and noise peaks

increase in the combined correlation output. Figure 3: FJTC correlation output of first channel.

To overcome these limitations, a filter is proposed

that minimizes the non-target peaks and noise as well

as equalizes non-equal target peaks if any. Figure 6

shows the correlation output after filtering where

distinct correlation peaks corresponding to the two

targets and reduced noise can be observed.

class-associative target detection processes using FAF

in terms of correlation peak intensity (CPI) and peak

to clutter ratio (PCR). PCR is defined as the ratio of

the desired target peak to the maximum non-target Figure 4: FJTC correlation output of second channel

peak. From table I it is obvious that the proposed

428

4. CONCLUSION

An improved multiple target detection technique

with better utilization of space bandwidth product is

proposed which employs different channel joint

transform correlation technique. In each channel one

member of the defined class is phase shifted and

phase-encoded to correlate with the target images.

Finally the different channel correlation outputs are

added to get the correlation plane output. This

technique produces a single correlation peak per

target and shows high discrimination for multiple

Figure 5: Combined correlation output. target detection even in noisy conditions by suitably

adjusting the FAF and the proposed filter

parameters.

5. ACKNOWLEDGMENT

This research is supported by Bangladesh University

of Engineering and Technology (BUET), Dhaka-

1000, Bangladesh.

after filtering.

associative target detection process

[1] A. B. VanderLugt, “Signal detection by complex

spatial filtering,” IEEE Trans. Inf. Theory, vol. IT,

Reference FJTC FJTC with post- no. 10, pp. 139-145, 1954.

images processing [2] C. S. Weaver and J. W. Goodman, “Technique for

used optically convolving two functions,” Appl. Opt., vol.

CPI PCR CPI PCR 5, p. 1248, 1966.

[3] M. R. Haider, M. N. Islam, M. S. Alam and J. Khan,

‘E’, ‘T’ 1,0.9753 1.5541 1,0.9999 2.4247 “Shifted phase-encoded fringe-adjusted joint

α=0.52, transform correlation for multiple target detection,”

β=0.48 Opt. Communi., 2004 (submitted).

[4] J. C. Kirsch and D. A. Gregory, “Video rate optical

‘E’, ‘M’ 1,0.9668 1.6411 1,0.9994 2.2695 correlation using a magneto-optic spatial light

α=0.52, modulator,” Opt. Eng., vol. 29, pp. 1122-1128, 1990.

β=0.48

‘E’, ‘Z’ 1,0.9838 1.7412 1,0.9999 2.4247

α=0.475,

β=0.525

429

- EE2170 Lab7 ManualUploaded byDiana Nahiely
- Band Stop Filter TodayUploaded byrehan
- Hairpin PejUploaded bynhilelan
- Robust switching vector median filter for impulsive noise removalUploaded byDaniel Perrefort
- HP-Bench-Briefs-1989-01-03Uploaded bychocobon_998078
- 12-235Uploaded bysumit_22in
- B_Tech_Civil_Engg.docxUploaded byravindrassk
- Blightertrack Fact Sheet Bss 1004Uploaded byannavikas
- 62 MicrowaveUploaded byNagaraju Neela
- Section 10 Two Variable Statistics (Workbook).pdfUploaded byMoonixey
- SBP Working Ppr 2017Uploaded bynaseem ashraf
- SSC CGL Tier II Study Schedule PDFUploaded byShrishanti Kale
- Atmaja_2016_J._Phys.__Conf._Ser._776_012072Uploaded byIbnu Sufajar
- woodUploaded byteckhuatl
- Definitions of Statistics TerminologyUploaded byFaridOraha
- The impacts of drought in agricultural productivity. An analysis at different scales for the two major rain-fed crops in Spain.Uploaded byMarina Svq
- Abstrak InggrisUploaded byEliatun Tarip

- Syllabus-3240Uploaded byCeh
- agilent LTEUploaded bymithunapurva
- Chapter08_OperatorOverloadingUploaded byMinhthức Bach
- und5_avaluosUploaded byDaima Nicola Bahir
- Ejercicios de fraccionesn 4° basico.docxUploaded byViviana Alexandra Ahumada Muñoz
- Plan Estrategico de MercadeoUploaded bylopezcla
- Cranes Dsp1 User Manual 6713Uploaded bytnj_dinesh
- Chap2Uploaded byAnonymous901
- 164-173Uploaded byPradeep_Vashist
- Guia2 VeranoUploaded byErnesto Lecuona
- Bend Tech Pro ManualUploaded byreaderofdata
- Rodrigues, R - Carta VII de PlatãoUploaded byTeotimo Lara
- Set 9 Reasoning Question PaperUploaded byPrince Kumar
- La Gerencia Educativa Desde Los Sistemas Adaptativos Complejos (Tesis Doctoral Mención Publicación) Rafael Perales l Unesr Feb 2014Uploaded byRafael Perales Leirós
- Aula 4Desenho Tecnico =)Uploaded byPaulo Sergio Martins
- Curva en r2 y en r3 Funcion Vectorial [Modo de Compatibilidad]Uploaded byoswaldo estrada bastidas
- material algebraUploaded bynando1953
- Kronecker productUploaded byrim_zakhama
- WEG Plc300 Comunicacao Canopen 10000849433 Manual Portugues BrUploaded byJosé Roberto Stanganini
- Raster Vector Data Analysis in DotSpatial(1)Uploaded bykaushik73
- RECTA PERPENDICULARES.docxUploaded byBrando Steven
- MeI - CódigosUploaded byAntonio Hurtado
- CH 22 MagnetismUploaded byMelinda
- conceptos ESTOCASTOCOSUploaded byKarina Storm
- prova de geometria descritiva.pdfUploaded byJose Bezerra
- SHS CURRICULUM.pptxUploaded byAdlih Liwen Bayang
- Calculo Integral Ing AmbientalUploaded byJuan Antonio Olivero Torres
- Torque GiroscópicoUploaded byIvan Garzon
- Glosario de Terminos OntologicosUploaded byPanem Et Circenses
- Modelling Twin Screw Multiphase Pumps a Realistic Approach to Determine the Entire Performance BehaviourUploaded byAnonymous Xy309m9Sm9