Iris Recognition Based On Wavelet Moment

Biometric Identification is a technology which takes the advantage ofhumans' physiology and behavior characteristic to carry on identification bycomputer. It comes over some short-comings in traditional identificationmethods. The operation of this technology is convenient and rapid. Because ofthe particularity of the iris structure, its venation is uniqueness and stability, IrisIdentification Technology is an important part of Biometric IdentificationTechnology and is considered as the most popular technology.PreprocessingBefore the process of feather extraction and recognition, we should carryon the preprocessing to the primitive iris image first, and this process concernsthe result directly. This paper has introduced classical Daugman algorithmwhich can locate the accurate margin of iris, then we locate the inside marginand outside margin of iris in two steps. The normalization of iris texture whichhas some bad influence brought by rotation, scaling and illumination variantshould be decreased to the same extent. The iris image can be mapped fromannular to rectangle of fixed dimension, then we do bilinear interpolation to thecoordinate of being mapped. At last, we enlarge the grade of grey level of theiris texture by methods of histogram equalization, more clear the detail is , moreeasily to recognize.Wavelet moment Wavelet moment is a kind of new-type moment which combines bothmoment and wavelet. Consider their each, wavelet moment not only has thecharacteristic of parallel invariance, scale invariance, rotate invariance andstrong noise immunity but also include multi-resolution characteristic ofwavelet. It has strengthened the ability to hold the exquisite characteristic of theimage by moment invariant.The general mathematic expression of moment can be described as follows:Fp q = ∫∫? f ( r , θ ) g p( r ) e jqθrdrdθWhere g p( r )is a function about variable r , p ,q ∈ Z.If g p( r )is a function defined in the global range of variable r , Fp q is aglobal characteristic. On the contrary, if g p( r ) is a function defined in localrange of variable r , then Fp qis a local characteristic.When { g p( r )}is chosen as wavelet basis function, we call Fp q as waveletmoment, and its sub-wavelet function system is:ψ a , b ( r )= 1aψ ??? r ?ab???Where a ( a ∈ R) is dilation factor, and b (b ∈ R)is displacement factor.The discrete form of dilation factor a and displacement factor b arechosen separately as a = a0m( m is an integer) and b = b0 a0m .Because of general normalization of size of the image being in the range{ r ≤ 1},we adopt the form of dyadic wavelet which is defined as:ψ m , n ( r ) = 2 m 2ψ(2 mr ? n)Let function do moment operation in all orientation, we get the global rangecharacteristic and local range characteristic based on variable m, n .we present the definition of wavelet moment invariant:1Fm w , anv , qel et2 = ∫0 S q ( r ) ?ψ m ,n( r )?rdr2Here m = 0,1,2,3. n = 0,1,2,3, ,2 m? 1 and q = 0,1,2,3.Feature extraction and pattern matchingAfter preprocessing, the iris image has become a rectangle regionD =[0,2π ] × [0,1], in the paper, it be partitioned into a series of equal sizerectangle: { Ai }1i=21 : 121iiA D=∪ =,and Ai ∩ Ai +1 ≠φ. We choose M=5 0 equaldistance points in [0,1] :Choose rectangle A1 = [0,2π ] × [0,5 ( M? 1)];A2 = [0,2π ] × [4 ( M ? 1),9 ( M? 1)],where the overlapped part A1 ∩ A2 = [0,2π ] × [4 ( M ? 1),5 ( M?1)];the rest may be deduced by analogy, rectangleA1 2 = [0,2π ] × [44 ( M ? 1),49 ( M? 1)],and its overlapped part of A1 1 is A1 1 ∩ A1 2 = [0,2π ] × [44 ( M ? 1),45 ( M?1)].Figure out wavelet moment vector's component of every rectangle Ai , asto rectangle A1 :Step 1. In the phase region {0 ≤ θ ≤ 2π}:2S q( r ) = ∫0 π f ( r , θ ) e jqθdθdisperse θ , ?θ = 2π / N, get one-dimension signal S q( r ) by algorithm FFT:10q( ) 2N( ,2 / )lS r f r l NNπ?π== ∑ exp ??? jqN 2πl???In order to utilize algorithm FFT, we should choose N as the form of theexponent of 2.Step 2. In the radial region of rectangle A1 :( 1 )5 ( 1)Fp qA = ∫0 M?S q ( r )ψ m ,n( r )rdrfix on sectional step-length 1?r = 49, the max value of radius is r0 = 5 ( M? 1),let Tr0r= ??? ???? , rk = k ?r :11( ), ,0( ) ( )Tw Am n q q k mn k kkF ?S r ψr r r== ∑ ?Here m = 0,1,2,3, n = 0,1,2,3, ,2 m? 1 and q = 0,1,2,3.We choose wavelet function as the cubic B-spline of Gauss approximationform :( )1 20 2( ) 4 cos 2 (2 1) exp(2 1)2 ( 1)2 ( 1)r a df rrψ = π d+ + ? σ ω ? π ? ? ??? ?σω?d+???here a = 0.697066, f 0= 0.409177, σω2= 0.561145, d= 3.Step 3. Calculate out wavelet moment vector:( )( 1 ) ( 1 ) ( 1 ) ( 1 ) ( 1 ) ( 1)φ ( A1 ) = F0 w,0 ,A0 2 , , F0 w,0 ,A3 2 , F1 ,w0 ,0A 2 , F1 ,w0 ,3A 2 , , F3 w,7 ,A0 2 , F3w,7 ,A32TAccording to this algorithm we can get wavelet moment vector:φ ( D ) = (φ ( A1 )T , , φ( A12) T )T.In order to reduce the dimensions of wavelet moment vector, the paper usea lower dimension subspace to describe iris image by Discrete CosineTransform(DCT).This paper use nearest center principle to classify and match . By the abovealgorithm, we extract 280 images belonging to 40 people from CASIA IrisDatabase,proved by the experiment, the algorithm of this article has gained ahigh recognition rate.