Expedient encoding system

Abstract

An encoding system for an iris recognition system. In particular, it presents a robust encoding method of the iris textures to compress the iris pixel information into few bits that constitute the iris barcode to be stored or matched against database templates of same form. The iris encoding system is relied on to extract key bits of information under various conditions of capture, such as illumination, obscuration or eye illuminations variations.

Description

BACKGROUND

The present invention pertains to recognition systems and particularly to biometric recognition systems. More particularly, the invention pertains to iris recognition systems.

Related applications may include U.S. patent application Ser. No. 10/979,129, filed Nov. 3, 2004, which is a continuation-in-part of U.S. patent application Ser. No. 10/655,124, filed Sep. 5, 2003; and U.S. patent application Ser. No. 11/382,373, filed May 9, 2006, which are hereby incorporated by reference.

U.S. Provisional Application No. 60/778,770, filed Mar. 3, 2006, is hereby incorporated by reference.

U.S. patent application Ser. No. 11/275,703, filed Jan. 25, 2006, is hereby incorporated by reference.

U.S. Provisional Application No. 60/647,270, filed Jan. 26, 2005, is hereby incorporated by reference.

U.S. patent application Ser. No. 11/043,366, filed Jan. 26, 2005, is hereby incorporated by reference.

U.S. patent application Ser. No. 11/372,854, filed Mar. 10, 2006, is hereby incorporated by reference.

U.S. patent application Ser. No. 11/672,108, filed Feb. 7, 2007, is hereby incorporated by reference.

U.S. patent application Ser. No. 11/675,424, filed Feb. 15, 2007 is hereby incorporated by reference.

U.S. patent application Ser. No. 11/681,614, filed Mar. 2, 2007 is hereby incorporated by reference.

SUMMARY

The present invention pertains to the iris recognition technology and human authentication methods. Iris patterns are proven to be unique and stable. The success of iris recognition system lies in using appropriate representations scheme of these unique iris patterns. This invention is about the representation of iris patterns extracted from the iris map. In particular, it presents a robust encoding method of the iris textures to compress the iris pixel information into few bits that constitute the iris barcode to be stored or matched against database templates of same form. The iris encoding method is reliable to extract key bits of information under various conditions of capture, such as illumination, obscuration or eye illuminations variations.

BRIEF DESCRIPTION OF THE DRAWING

FIG. 1 is a diagram of an overall iris recognition system;

FIG. 2 is a diagram of waveforms that may be used relative to a filter in conjunction with an encoding scheme;

FIG. 3 is a diagram showing a construction of the even odd components of a signal;

FIG. 4 shows a basic mask and barcode layout for an encoding algorithm;

FIG. 5 is a diagram of an encoding approach using convolution;

FIG. 6 is a diagram of an encoding approach using a dot product;

FIG. 7 is a diagram of an encoding scheme using a binning approach; and

FIGS. 8 a and 8b are diagrams illustrating an encoding scheme with binning of a barcode based on minima and maxima of signals.

DESCRIPTION

The present system may relate to biometrics, an iris recognition system, image metrics, authentication, access control, monitoring, identification, and security and surveillance systems. The present system addresses processing procedure of iris encoding to support in development of improved iris recognition systems. The present system may provide methods to compress the extracted normalized iris map image into a compact bit representation of the iris pixels while preserving the key iris pattern information. This compact representation of iris may be computed to execute an accurate matching and enrollment.

The overall eye detection system is shown in FIG. 1. It shows a digital camera 61 that may provide an image with a face in it to the eye finder 62 as noted herein. The eyefinder 62 may provide an image of one or two eyes that go to the iris segmentation block 63. A polar segmentation (POSE) system in block 63 may be used to perform the segmentation. POSE may be based on the assumption that image (e.g., 320×240 pixels) has a visible pupil where iris can be partially visible. There may be pupil segmentation at the inner border between the iris and pupil and segmentation at the outer border between the iris and the sclera and iris and eyelids. An output having a segmented image may go to a block 64 for mapping/normalization and feature extraction. An output from block 64 may go to an encoding block 65 which may provide an output, such as a barcode of the images to block put in terms of ones and zeros. The coding of the images may provide a basis for storage in block 66 of the eye information which may be used for enrolling, indexing, matching, and so on, at block 67, of the eye information, such as that of the iris and pupil, related to the eye.

One may extract and encode the most discriminating information present in an iris pattern. Just significant features of the iris texture are encoded so that comparisons between templates may be made faster and more reliable. Many iris recognition systems might use a band pass decomposition of the iris image using two-dimensional (2D) modulated filters with multiple parameter dependencies. A present simplified 1D phase-based encoding approach may use a single periodic filter configured by a single parameter.

The present approach may be staged into multiple steps to extract features at different central frequencies and at different phasor quantizations. The approach may compress an iris pattern into fewer bits to an extent to make a match without having to compute all bits (minimum savings may reach fifty percent relative to the bit count of other approaches), thus allowing for efficient storage and fast comparison of large iris databases. In addition, the present encoder may be an extension to what is implemented to segment the iris boundaries, and thus some of the approach may be executed at an early stage during segmentation to save on the computation load.

A key component of iris recognition system may be an encoding scheme to extract the key features of the iris texture into fewer bits which are then used to match the subjects. The matching may be significantly influenced by many factors including the segmentation, feature extraction, and spatial resolution and image quality.

The present approach may extract and encode the most discriminating information present in an iris pattern. The present feature extraction and encoding scheme may be embedded within one-dimensional polar segmentation (1D POSE) and thus reduce sources of errors and allow for staged matching for fast iris indexing. Just the significant features of the iris texture are to be encoded so that comparisons between templates may be made unbiased. Many other iris recognition systems may use a band pass decomposition of the iris image using a 2D Gabor (i.e., a modulated sine and cosine waves with a Gaussian function) or in general wavelet functions to create a biometric template.

Wavelets may be used to decompose the data in the iris map into components that presented at different levels of resolution. A number of wavelet filters may often be applied to a 2D iris map at multi-resolution levels to extract localized features in spectral and spatial domains, and allow matching at multilevel resolutions.

Gabor or log-Gabor filters (somewhat popular in iris recognition systems) appear to be simply subsets of wavelet functions and may be able to provide conjoint representations of the features in space and spatial frequency. These techniques might be effective in extracting the iris features; however, their implementation and configurations appear to involve multiple parameter settings and other computational burdens. While the current implementations of iris encoding may be well represented by any wavelet function or 2D Gabor functions, the Gabor decomposition is difficult to compute and lacks some of the mathematical conveniences that are desired for good implementations, such as not being invertible, non-linear reconstruction, and maltreatment of DC components.

The present encoder may incorporate 1D encoding, a single filter to extract phase information, and a simple unbiased filter to cancel out DC components and extract just significant discriminating information present in the phase content of an iris pattern. The encoder may compress the iris pattern into fewer bits (e.g., the iris code may use just one-half bit counts of current iris code methods. The encoder approach may be staged into multiple independent steps thus allowing flexibility in producing code bits to an extent to make a match without having to compute all of the bits, and some of the encoder approach steps may be executed at an early stage during segmentation to save on the computation load.

The present approach and system may extract and encode the most discriminating information present in an iris pattern using a straightforward approach that extends upon the Haar wavelet filters to any form of a symmetric waveform. The approach may start with a feature vector, i.e. intensity function as a function of radial variable extracted from the iris image at each angle. The feature vector may be interpolated to generate a radial resolution covering the features of the iris pattern between the two iris borders at specified angles. Then one may dot product the extracted feature vector by a single periodic filter. Various waveforms may be used to construct the filter with an emphasis that the symmetric waveform sums to zero to cancel out any DC components and eliminate unbiased results (e.g., a Gabor filter may suffer from this kind of bias). One may then capture the phase content of the feature vector by computing the sum over a shifted segment/window (i.e., window width equals the waveform time period) corresponding to the selected center frequency. Thus, an iris template may be generated as a compressed version of the generated phasor feature elements. The summed feature vector elements may be sign quantized so that a positive value is represented as 1, and a negative value as 0 (or vice versa). This may result in a compact biometric template consisting of half counts of bits of related art encoding approaches. Additional bits may also be generated by repeating the same procedure using shifted versions of the filter.

The present approach may start with the feature vector extracted from the iris image at each angle. The feature vector may be interpolated to generate the radial resolution covering the features of the iris pattern between the two iris borders at the specified angle. Then one may dot product the extracted feature vector by a single periodic filter. Various waveforms may be used to construct the filter with an emphasis that the symmetric waveform sums to zero to cancel out any DC components and eliminate unbiased results. Then one may capture the phase content of the feature vector by computing the sum over a shifted segment/window (window width equals to the waveform time period) corresponding to the selected center frequency. Thus, an iris template may be generated as a compressed version of these generated phasor feature elements. The summed feature vector elements may be sign quantized so that a significant positive value is represented as 1, a significant negative value as 0, and insignificant value close to zero is defined by an unknown bit as an x. This may result into a more accurate presentation of the iris patterns by excluding the uncertain bits associated with noise and interference distortions. Additional bits may also be generated by repeating the same procedure using shifted versions of the filter.

To compress an image, one may encode it. The may be a map, having radial resolution versus angular resolution. The radial resolution (R_R) may have a 100 points and the angular resolution (A_R) may have 360 degrees. However, one may do just every other degree going completely around the eye to end up with 180 degrees for the angular resolution. The map may be of an iris. The radial marks on the iris may be decimated or interpolated.

The size of the data may be R_R×A_R bytes. Each pixel may be a byte with, for instance, 8 bits per byte. A goal of the encoding may be to compress data down to small quantity or size. The present invention or approach could take the image and run it through a log Gabor wavelet to result in a compressed image with a sign of two outputs—real and imaginary which may be done in the related art. Unlike that art, the present invention may do the analysis or encoding just on the radial axis for each angle. One may have radial values extracted at each angle and at the specific radial resolution. Encoding may be done at the same time as the segmentation. One may have three outcomes (i.e., 0, 1 and unknown (x) for bit representation. The related art may just have two outcomes, 1 and 0, and assign a value of 1 or 0, which is a strong indication for a weak representation of values that at the transition from positive to negative or vice versa. The present system may realistically place a value, as it appears, which is a 1, 0, or an unknown x for the insignificant values approaching zero. It may be better to note just the strong values and ignore the insignificant as signs can vary dramatically at values close to zero. Further, the present encoding scheme may deal with just one dimension and not two. The information here may be on a string or the radial of the material to be mapped and encoded.

Iris encoding may be a key component of an iris recognition system and may be used to extract the key features of the iris texture into fewer bits which are then used to match the subjects. The matching may be significantly influenced by many factors including the segmentation, feature extraction, spatial resolution and image quality. The present approach may extract and encode the most discriminating information present in an iris pattern. The present feature extraction and encoding scheme may be embedded within the 1D POSE segmentation to reduce sources of errors and allow for staged matching for fast iris indexing. With the present approach, just the significant features of the iris texture are to be encoded so that comparisons between templates may be made fast and unbiased.

The present approach may be based on a 1D analysis. A 1D feature may be advantageous over the 2D feature extraction in terms of computation and robustness. To avoid unbiased features, one may convolve the filters only in the radial axis. Convolving the filters in both directions at different scales may degrade performance. The radial direction may have most of the crucial information and preserve iris information regardless whether the pupil is dilated or not. On the other hand, convolving the filters on the angular direction may be affected by the occlusions of the iris as well as by the under-sampling of the iris map. The present system may deploy a one-dimensional approach applied in a radial direction. As part of the POSE segmentation technique, the intensities may be convolved by a step function. So the feature extraction may be may be combined with the segmentation into a single step and thus reduce computation.

Decomposition of the intensity values may be accomplished by using a new set of filters as an extension to 1D Haar wavelet functions or step function where the emphasis is made to construct a periodic one to extract quadratic information of the intensity variations. The filter outputs may then be binarized based on the sign values to present the real and imaginary parts of the equivalent Gabor filters without constructing Gabor filters.

The present approach may be staged into multiple steps that permit fast and quick matches without processing the entire iris code. The approach may allow extracting additional iris code to characterize the iris texture at different central frequencies and at a higher order complex domain. Extraction of additional bits may be possible with the present approach and may be staged as needed if a match does not occur at the first settings.

Instead of constructing two filters, one may construct one 1D periodic filter (note FIG. 2 which shows filter samples 11 to compute the quadratic phase information of the signal) as an extension to a Haar wavelet or step function, sine wave function or any other form of smooth symmetric functions. The filter may be constructed to have zero sums over its period in order to eliminate DC component accumulations and just phase information will be maintained. Taking just the phase may allow encoding of discriminating textures in two irises, while discarding effects due to illumination variations which are inherited in the amplitude information. Phase information rather than amplitude information may provide reliable characteristics of digital images. The amplitude component may often represent the illumination variation and be heavily affected by noise.

Decomposition of the intensity signal may be accomplished with the following items. Convolution may be effected using a filter of a single period wave and having a central frequency specified by the period T. The dot product of the signal and the filter may be constructed to generate an output signal,y(r)=I_θ(r)·ƒ(r).The signal I_θ(r) may denote the intensity signal extracted at each angle as a function of the radius values. These may be interpolated image intensity values between the two boundaries of the iris and be sampled to a predefined number of radius samples N_r. The function ƒ(r) may represent the filter function of length N_r.

One may sum over a “T” period 12 of the output signal 13 using a single, two shifted sum 14 (that constitutes the even and odd symmetry components of the intensity signal) or be even more based upon multiple shifts as shown in FIG. 3. This process may result into two output signals, i.e., if two shifts deployed, as a function of just the phase information while amplitude information is being canceled because of the nature of the symmetry inherited in the filter signal. FIG. 3 is a diagram showing a construction of the even odd components of the signal in a single step. The shifting ΔT 14 may be any fraction of the period T 12. To match the quadratic output of a Haar Wavelet or Gabor quadratic imaginary and real outputs, one may set it to half of the period T, ΔT=T/2. A sign function may then be used to binarize the filtered values thus present the two possible values by a 0 or 1 bit, resulting into four possible quadratic values equivalent to real and imaginary components constructed by a Gabor filter output.

Unlike the Gabor or wavelet approach, the present approach may allow splitting the outputs into two or more stages for quick indexing. Since the quadratic information may be computed using separate functions, one can stage the encoding approach using first the non-shifted function and then computing an iris code having the same size as the iris map, i.e., N_r×N_θ. The mask matrix may also be constructed at this smaller size than previously done. There appears to be no need to duplicate the mask size as done in a known Daugman encoding approach. Unlike a related art approach, one may use the present approach to extract additional codes as needed based upon filter outputs of different period shifts, as well as scaled periods for different central frequencies if a match does not occur. The approach may provide flexibility to stage the matching process and allow extraction of fewer bits to make a match.

Encoding may be a way to compress the most discriminating information present within the iris map into fewer bits so that comparisons between templates can be made real-time. One may make use of multi-band decomposition of the iris map to extract the fine and coarse information content of the iris distinctive patterns. A present method for iris feature encoding may be presented in several algorithms.

Wavelets may be used to decompose the iris map into bank of filters to extract wavelet coefficients at different resolutions. Wavelet coefficients may then be encoded at each band to compress the map into fewer bits representing the iris signature. An advantage of using wavelet is that it may be well localized in both spatial and frequency domain.

As to the Gabor/Log Gabor wavelet, Daugman appeared to make use of a two-dimensional (2D) Gabor filter to encode iris maps. A Gabor filter may be built on the basis of sine cosine wave modulation with a Gaussian waveform. This may make it as a special case of a wavelet and thus it can indeed localize features in both spatial and frequency domains.

Decomposition of an image may be accomplished using a quadrature pair of Gabor filters with real parts associated with the cosine modulation and the imaginary part associated with the sine modulation. The sign of the real and imaginary parts may be used to quantize the phase information into four levels using 0/1 bits for positive/negative signs of each of the real and imaginary components.

The Haar wavelet may be a simplified version of a wavelet transform to extract features from the iris map. Gabor and a like wavelet may require many parameters for setting and configuration.

In the options, four levels may be represented using the two bits of data, so each pixel in the iris map corresponds to two bits of data in the iris barcode (template). A total of Nr×Nq×2×L bits may be calculated for each barcode. L=number of bands, Nr and Nq indicate the size of the iris map.

The present encoding scheme may be applied to a 1D signal using radial signal rather than a 2D map. A three bit representation may be used rather than a two bit representation. One may extract as many bits (i.e., blocks of bits) as needed to quantify the information in the phasor only (i.e., no amplitude). This may be important when there are limited iris region due to obscuration (fewer bin comparisons). Thus, the more bit blocks that are extracted, then better matching may be expected. In addition, one may choose to use simple form waveforms (as shown in FIG. 2) which may be simpler designs than Gabor or any wavelet forms with only period as a single parameter.

Three-bit representation may be used in the present schemes. In other approaches, the feature vector may be sign quantized so that any positive value is represented by 1, and negative value by 0. However, in the present approach, the quantization may use three levels, in that a positive value is represented by 1, a negative value is represented by 0, and a value close to zero, i.e., ≦ν (tolerance), is represented by x (unknown). An unknown bit may be either 0 or 1.

The present approach and system may utilize various schemes of encoding. One scheme may be like the related art except that one may modify its technical approach to include three outcomes from the encoder rather than two outcomes as in the related art. The present encoding scheme may be applied to a one-dimensional (1D) signal using a radial signal rather than a 2D map.

The iris encoding scheme may have options of waveforms. Using one of the present algorithms, one may extract as many bit blocks as wished based on the following variations of the first, second and third algorithms, respectively. For any period selection, one may obtain a new set of bits; for any shift of the period, one may obtain a new set of bits; and for any wavelength, one may obtain a new set of bits. A fourth algorithm may result into a single block of bits. FIG. 4 is a diagram showing a basic mask and barcode layout applicable for some of the algorithms. The layout may include a set of masks 41, 42, 43 and 44 and corresponding barcodes 46, 47, 48 and 49. Waveform information, relative to the masks and barcodes 1, 2, 3 . . . k, is indicated by T₁, ΔT₁, u₁(r), T₁, ΔT₂, u₁(r), T₂, ΔT₁, u₁(r), . . . T₁, ΔT₁, u₂(r), respectively.

A signal may be convoluted. One may get scores for each value of the waveform. A convolution result or output may be f(r). If f(r) is greater than gamma (γ), then it may be one; if it is less than gamma, then it may be zero; and if it is within gamma or less than gamma, then it may be unknown (x). An unknown measure of a pixel may be masked as the masked information may be unknown. Again, this analysis may be done just on the radial one dimension.

One may run a sign test and end up with just one bit per value and save 50 percent on a use of bits. If the signal is not sufficiently discriminant for a match, then one may do a shift to get another bit. Convolution may be done on a shifted version of the same wavelength. Shift ΔT may equal T/2.

A general form to convolve with a shifted version of the same waveform may bef(r_k)=I(r)*u(r−ΔT_k)where I(r) is an intensity vector and * is the convolve symbol.

A goal is to have one bit, but if one does not get a match, one may increase it to two, three or more bits until a match is obtained. In a closed form, the unknown notion of 1, 0, x, may be used for an outcome. The complete period of a signal may be used. For each shifting, the sign test may be performed. The waveform may be generalized.

FIG. 5 is a diagram of the encoding approach using a first algorithm. One may base the iris encoding on a single waveform filter which is either odd or even symmetric. Then one may make use of circular shifted version of the waveform to obtain additional contents of the data. Rather than using Gabor, may make use of the step function as the symmetric waveform to be convolved with 1D radial signal. Unlike Gabor, (often used in existing art), one may use step symmetric waveform because it cancel out the DC components. Mathematically, this may be deduced to simple difference among integrals of the actual values of the intensity function. For example, using one shift of half period, the formulation can be simplified as noted in the following.

f(r)=I(r)*u(r), where f(r) may be a result of a convolution, I(r) is an operator and * indicates convolving. u(r) may indicate for an example a step function 51 as shown in FIG. 5. The following may lead to a sign test.

$\Rightarrow f\left(r\right)={\int }_{-T/2}^{T/2}I\left(r\right)dr-{\int }_{-T}^{-T/2}I\left(r\right)dr-{\int }_{+T/2}^{T}I\left(r\right)dr<>\gamma \Rightarrow b\left(r\right)=\{\begin{array}{cccc}=1& \mathrm{if}& f\left(r\right)& >>\gamma \\ =0& \mathrm{if}& f\left(r\right)& <<\mathrm{-\gamma }\\ =x& \mathrm{if}& f\left(r\right)& \approx \mathrm{or}<\gamma \end{array}$ Note that “x” means “unknown” which means that the pixel has to be masked in the barcode and it is not relevant to set it to either 0 or 1. The same applies to a shifted version of u(r).

For ΔT=T/2, one may have

$f\left(r\right)={\int }_0^{T}I\left(r\right)dr-{\int }_{-T}^{0}I\left(r\right)dr<>\gamma .$

FIG. 5 shows a diagram with a step function input 51 to a block 52, which has an output to block 53, which in turn has an output to block 54. Blocks 52, 53 and 54 represent u(t−ΔT₁), u(t−ΔT₂), and u(t−ΔT_k), respectively. “k” may be 3 or some other number according to how many of the u(t−ΔT_k) blocks may be had. This approach using “k” may be application to other components and/or symbols of the present Figure and the other Figures referred to herein. Outputs from blocks 52, 53 and 54 may go to convolution operators 99, 56 and 57, respectively. Also input to each of the operators may be an output from the iris map 55. Outputs from operators 99, 56 and 57 may go to blocks 58, 59 and 71, respectively, each representing (|f(r)|−ν). The outputs from operators 99, 56 and 57 may also go to diamond symbols 76, 77 and 77, respectively, that ask a question, “sign (f(r))<0?”. Outputs from blocks 58, 59 and 71 may go to diamond symbols 72, 73 and 74, respectively, which ask the question, “≦0?” Map mask 79 may have an output to a diamond symbol 75, which asks the question, “mask_θ(r)>0?” A “yes” answer from diamond 72, 73 or 74 may go as a one to a code mask 41, 42 or 44, respectively. A “no” answers from one of diamond symbol 72, 73 or 74 may go to the diamond symbol 75. If an answer from symbol 75 is yes, then a one may go to code mask 41, 42 or 44, respectively. If an answer from symbol 75 is no, then a zero may go to code mask 41, 42 or 44, respectively. With respect to diamond symbol 76, 77 or 78, if an answer is yes, then a zero may go to the barcode 46, 47 or 49, respectively. If an answer from symbol 76, 77 or 78 is no, then a one may go to the barcode 46, 47 or 49, respectively.

Another approach or scheme of encoding would not use convolution of FIG. 5 but a dot product as indicated in FIG. 6. One may take the wavelength and the dot product and use the sign test. To convolve is not needed in the version of FIG. 6. The same test may be used to come up with the three bits. A shift may be done to get more bits. More bins may be obtained for all T, to get more global stretch wavelength, wider, and get a new set of codes. One may stretch or squeeze. For a new wavelength, a different set of codes may be obtained. For all T one may get different codes for different frequencies. The algorithm deploys a dot product on the signal with a periodic filter.

$f\left(r\right)={\int }_{-T/2}^{T/2}I\left(r\right)·u\left(r\right)dr-{\int }_{-T}^{-T/2}I\left(r\right)·u\left(r\right)dr-{\int }_{+T/2}^{T}I\left(r\right)·u\left(r\right)dr<>\gamma \Rightarrow b\left(r\right)\{\begin{array}{ccc}=1& \leftarrow f\left(r\right)& >>\gamma \\ =0& f\left(r\right)& <<\mathrm{-\gamma }\\ =x& f\left(r\right)& <\mathrm{or}\approx \gamma \end{array}$ The preceding may be regarded as a sign test for a bit. The following might be noted.

• • ∀ΔT, one may obtain a new set of bins based upon demand. It is probably adequate to run the scheme algorithm only once to make a match.
  • ∀T₁, one may obtain a new set of codes based upon frequency content.
  • ∀ wavelength, one may obtain a new set of codes based upon the peaks attenuation.

FIG. 6 shows a diagram with a square waveform (having a somewhat constant period) input 51 to a block 52, which has an output to block 53, which in turn has an output to block 54. u(t) may have the same length as a radial axis. Blocks 52, 53 and 54 represent u(t−ΔT₁), u(t−ΔT₂), and u(t−ΔT_k), respectively. “k” may be 3 or some other number according to how many of the u(t−ΔT_k) blocks may be had. This approach using “k” may be application to other components and/or symbols of the present Figure and the other Figures referred to herein. Outputs from blocks 52, 53 and 54 may go to dot product operators 82, 83 and 84, respectively. Also input to each of the operators may be an output from the iris map 55. Outputs from operators 82, 83 and 84 may go to blocks 58, 59 and 71, respectively, each representing (|f(r)|−ν). The outputs from operators 82, 83 and 84 may also go to diamond symbols 76, 77 and 77, respectively, that ask a question, “sign (f(r))<0?”. Outputs from blocks 58, 59 and 71 may go to diamond symbols 72, 73 and 74, respectively, which ask the question, “≦0?”. Map mask 79 may have an output to a diamond symbol 75, which asks the question, “mask_θ(r)>0?”. A “yes” answer from diamond 72, 73 or 74 may go as a one to a code mask 41, 42 or 44, respectively. A “no” answers from one of diamond symbol 72, 73 or 74 may go to the diamond symbol 75. If an answer from symbol 75 is yes, then a one may go to code mask 41, 42 or 44, respectively. If an answer from symbol 75 is no, then a zero may go to code mask 41, 42 or 44, respectively. With respect to diamond symbol 76, 77 or 78, if an answer is yes, then a zero may go to the barcode 46, 47 or 49, respectively. If an answer from symbol 76, 77 or 78 is no, then a one may go to the barcode 46, 47 or 49, respectively.

In another approach or scheme, one might “bin it”, having a waveform 91 as shown in FIG. 7, which may reveal non-uniform binning. Smaller bins may be made at the inner bounds of the iris because there is more information closer to the pupil. Thus, a smaller bin may be utilized. This approach may be tied into an amount of coverage. An integral may be used starting from a given period to a certain T, and then the difference be compared of earlier on. A sign of being bigger may be a 1, being smaller a 0, or not known an x.

The scheme may be based on bins (i.e., binning approach) which determine the localized features of a signal within the bins. The bins may be shown to have a length T. The bins do not necessarily have to be uniform. For instance, one may chose to have smaller bins at the vicinity of the inner bound and larger bins at the outer bound (where the SNR is expected to be smaller).

FIG. 7 shows a diagram with a square waveform (having a varying period) input 91 to a block 52, which has an output to block 53, which in turn has an output to block 54. u(t) may have the same length as a radial axis. Blocks 52, 53 and 54 represent u(t−ΔT₁), u(t−ΔT₂), and u(t−ΔT_k), respectively. “k” may be 3 or some other number according to how many of the u(t−ΔT_k) blocks may be had. This approach using “k” may be application to other components and/or symbols of the present Figure and the other Figures referred to herein. Outputs from blocks 52, 53 and 54 may go to dot product operators 82, 83 and 84, respectively. Also input to each of the operators may be an output from the iris map 55. Outputs from operators 82, 83 and 84 may go to blocks 58, 59 and 71, respectively, each representing (|f(r)|−ν). The outputs from operators 82, 83 and 84 may also go to diamond symbols 76, 77 and 77, respectively, that ask a question, “sign (f(r))<0?”. Outputs from blocks 58, 59 and 71 may go to diamond symbols 72, 73 and 74, respectively, which ask the question, “≦0?”. Map mask 79 may have an output to a diamond symbol 75, which asks the question, “mask_θ(r)>0?”. A “yes” answer from diamond 72, 73 or 74 may go as a one to a code mask 41, 42 or 44, respectively. A “no” answers from one of diamond symbol 72, 73 or 74 may go to the diamond symbol 75. If an answer from symbol 75 is yes, then a one may go to code mask 41, 42 or 44, respectively. If an answer from symbol 75 is no, then a zero may go to code mask 41, 42 or 44, respectively. With respect to diamond symbol 76, 77 or 78, if an answer is yes, then a zero may go to the barcode 46, 47 or 49, respectively. If an answer from symbol 76, 77 or 78 is no, then a one may go to the barcode 46, 47 or 49, respectively.

In another approach or scheme, as indicated in FIGS. 8a and 8b, one may have

∀f(r)=I(r)−ΣI(r) within the two valleys. In a sense,

$\forall f\left(r\right)=I\left(r\right)-\sum _{\underset{\mathrm{valleys}}{\mathrm{within}}}I\left(r\right)<>\gamma .$ There may be a move to capture the peaks and valleys, and use 1's for peaks and 0's for valleys with respect to average values. About everything else may be regarded as unknown, i.e., transition areas where f(r) approaches zero.

The binning of the barcode may be based upon the local minima 92 and maxima 93 of the radial signal per each angle, as shown in a graph 94 of FIG. 8a. Also, the values 95 in each bin are indicated as 1 (one), 0 (zero) and x (unknown). Several criteria may be noted in the following.

$\begin{array}{cc}f\left(r\right)=I\left(r\right)-\frac{1}{\left(M_{k+1}-M_k\right)}\sum _{r\in \left[M_k,M_{k+1}\right]}I\left(r\right)\Rightarrow b\left(r\right)=\begin{array}{cc}0;& \mathrm{if}f\left(r\right)<0\\ x;& \mathrm{if}f\left(r\right)\ge 0\end{array}& \left(101\right)\\ f\left(r\right)=I\left(r\right)-\frac{1}{\left(m_{k+1}-m_k\right)}\sum _{r\in \left[m_k,m_{k+1}\right]}I\left(r\right)\Rightarrow b\left(r\right)=\begin{array}{cc}1;& \mathrm{if}f\left(r\right)>0\\ x;& \mathrm{if}f\left(r\right)\le 0\end{array}& \left(102\right)\end{array}$ When there are two outcomes for the same pixel, the confirmed bits may be selected over the unknown bit choice.

FIG. 8 b is a diagram of another algorithm noted herein. An iris map 96 which has its peaks 97 and valleys 98 located. These located peaks and valleys and other map 96 information may have equations 101 and 102, respectively, applied to them. The results from equations 101 and 102 may go to diamond symbols 103 and 104, which asks a question, “sign (f(r))<0?” A map mask 105, corresponding to iris map 96, may have an output to a diamond symbol 106, which asks a question, “mask_θ(r)>0?” If an answer to the question of symbol 106 is yes, then a one may go to a code mask 108, and if the answer is no, then a zero may go to the code mask 108. If an answer to the question of symbol 103 is yes, then a zero may go to a barcode 107 and symbol 106, and if the answer is no, then a one may go to the code mask 108. If an answer to the question of symbol 104 is yes, then a one may go the barcode 107 and the symbol 106, and if the answer is no, then a zero may go to the code mask 108.

An analysis (i.e., encoding) may be performed on the radial axis per each angle.f(r)=I(r)*u(r)

Sign Test γ

0/1 or x.

To obtain additional bits per each pixel value,f(r)=I(r)*u(r−ΔT_k).

The three bit approach may be regarded as a trick to eliminate much noise. It may be good for f(r) values as they are considered as unknown since a value is not assigned to it. As to an unknown, a separate weight may be assigned. The weight may vary between from low to high but not be a 100 percent of either extreme since that would amount to one of the other two values. This weighting approach may handle the encoding uncertainty or noise but not the segmentation noise.

In the present specification, some of the matter may be of a hypothetical or prophetic nature although stated in another manner or tense.

Although the invention has been described with respect to at least one illustrative example, many variations and modifications will become apparent to those skilled in the art upon reading the present specification. It is therefore the intention that the appended claims be interpreted as broadly as possible in view of the prior art to include all such variations and modifications.

Claims

1. A method for obtaining an image of an iris and encoding pixel information contained within the image, comprising a surveillance system to perform: capturing an image of an iris with a digital camera;employing an iris recognition system to segment the captured image, wherein segmentation isolates an annular iris from the remainder of the captured image;employing the recognition system to extract from the segmented image of iris intensity vectors of the image at each of several angles,applying a periodic filter to the intensity vectors by employing the recognition system to apply a periodic filter to the intensity vectors thereby encoding the filtered intensity vectors;representing the encoding of the filtered intensity vectors as a barcode; andstoring the representation of the barcode in a non-transient storage medium.

2. The iris encoding method of claim 1, wherein the periodic filter is constructed with a symmetric waveform.

3. The iris encoding method of claim 2, wherein the symmetric waveform sums to zero over its period.

4. The iris encoding method of claim 1, wherein each feature vector may be sign quantized with a one, zero, or unknown.

5. The iris encoding method of claim 1, wherein the feature vector from the iris image at each of several angles is encoded into fewer bits to be used for matching.

6. The iris encoding method of claim 1, wherein the feature vector from the iris image at each of several angles is encoded in stages into fewer bits to be used for matching.

7. The iris encoding method as claimed in any preceding claim, wherein extracting the encoding step is integrated with segmenting the image.