Human body trick recognition method, is also called the biological features recognition methods, is refers to using person’s unique physiology and the behavior characteristic carries on the distinction the identification authentication technological means. Its production and the development stem from the people to make great strides forward the Digital Age in the process the demand which enhances unceasingly to the identification authentication method’s accuracy and convenient. The traditional identification authentication method mainly includes the status designated object (for example key, credential and so on) as well as the status symbol information (for example account number, password and so on), or above the two’s union (for example bankcard and so on). The people discovered in the use process that they have the common shortcoming: Easy to lose and to fabricate. Moreover the traditional identification authentication system cannot the effective recognition have these status symbol thing person whether is the genuine owner. Therefore, once pretends to be, the genuine owner will suffer the enormous loss. Therefore, the human body trick recognition method takes a more effective solution to obtain the widespread application gradually.
The human body characteristic’s distinction method has many kinds. In all biological features, fingerprint relative stabilization, but the enrollment fingerprint is not non-offensive. The face characteristic has many merits likely (for example initiative, non-offensive and user friendly and so on), but the face changes likely along with the age, moreover is camouflaged easily. The sound characteristic has with the face likely characteristic similar merit, but it along with factors and so on age, state of health and environment changes, moreover the storyteller recognition system easily is also recorded deceives, is fabricated easily. The iris trick recognition has solved these problems, but also has some merits which the above other biological features do not have, therefore the recent years iris recognition technology was considered that is one of most promising biological recognition technologies.
Iris recognition technology general process
The iris recognition technology’s process generally speaking divides into: The iris image gain, the image pretreatment, the feature extraction and the characteristic match four steps.
The iris image gain is refers to the use specific digit photograph equipment to carry on the photography to person’s entire ocular region, and will photograph the image transmits in the computer through the image gathering card to save.
The image pretreatment is refers to, because photographs the ocular region image has included the unnecessary information, and in aspects and so on clarity cannot satisfy the request, needs to carry on to it including pretreatment operations and so on image smoothing, marginal check, image separation.
The feature extraction is refers to through certain algorithm from the iris image which separates withdraws the unique characteristic point, and carries on the code to it.
Finally, the characteristic match is refers to the iris image feature coding which in the basis feature coding and the database saves beforehand to carry on compares to, the confirmation, thus achieves the recognition the goal.
Gain ocular region image
This article iris image ingestion installment as shown in Figure 1, what uses is Zhuo Wei (SOVIC) the SP-313 camera. What this camera uses is newest CCD the effect CMOS photosensitive chip, the image resolution is 350,000 picture elements (640×480 non-software interpolation), the built-in low light intensity’s auxiliary photo source, can maximum limit reduce to the human eye stimulation, when the use matches by the artificial darkroom, causes person’s ocular region image to be clearer brightly. The person ocular region image which Figure 2 is this design uses the camera which gains.
Figure 1 iris image ingestion installment
Gains after the picture data, only needs it to read in according to certain picture format the document, then completes the ocular region image in computer memory which needs. In what this article procedure uses is the BMP form image document, because the BMP image file memory’s image data has not undergone the compression, will facilitate later the pretreatment which will carry on to the image.
Figure 2 person of ocular region image
Ocular region image pretreatment
The BMP image document format mainly has 1, 4, 8, 16, 24 and 32 and so on image forms. 32 BMP image document format expressed that this image has 232 kind of colors, in image each picture element with 32 expressions, in the ordinary circumstances this document format has not adjusted the color print, 32 highest 8 retentions, other 8 expressions are red, 8 expression greens, 8 expression blue color. 8 BMP image document expressed that this image has 256 kind of colors. In image each picture element with 8 expressions, and searches this picture element with these 8 achievement index in the colored table the color, 8 BMP images are also called the gradation image generally.
The image which gains in this article is 32 colored BMP images. 32 color images save the image color data are many, the image document’s size is also big. But looking from this article pattern recognition’s request, these is nonessential, therefore has the necessity its transformation is 8 gradation images.
The conversion formula like type (1) shows.
(1)
And Gray (i, after j) is the transformation black-and-white image in (i, j) place grey level, because in the formula the green occupies the proportion is biggest, therefore time transformation may after meets uses the G value takes the transformation the gradation. After transformation gradation image as shown in Figure 3. Looks from the image with 32 RGB image not big difference, but the image document’s size reduced from 1.17Mb to 301Kb.
Figure 3 person of ocular region image gradation image
Will gain the ocular region image transforms after the gradation image, but also needs to carry on the denoising acoustic treatment to the gradation image. What this article uses is in the air zone law weighting average value filter, it has the odd point sliding window with one to skid on the image, replaces a window center point correspondence’s image picture element grey level with the window in each spot grey level’s mean value, if the sliding window had stipulated in takes the weight which in the average value process a window each picture element institute occupies, is also each picture element coefficient.
Extraction iris image
This process needs to read the ocular region image the data, examines the iris image the inside and outside edges, the extraction internal bore center of circle coordinate and the semi-minor axis, extracts the iris long radius again, establishes the polar coordinate system, separates the iris image, finally carries on the feature extraction.
Compare with eye’s other parts, pupil’s grey level must be much smaller, is also the color must be much darker, and has an obvious sudden change on the gradation level, i.e. must “be much blacker than” in pupil’s gradation level other part of gradation level. Therefore, may use this characteristic fully, carries on the histogram analysis to Figure 2, result as shown in Figure 4.
Figure 4 gradation histogram
The computed result may obtain to Figure 4, image grey level from 62 starts, and in the chart has certain peak points. Our known pupil’s color is darkest, therefore may determine the first wave ridge as pupil’s gradation distribution. Observes the first peak value specifically, it assumes the sine function shape distribution basically, take 72 as wave ridge (value: 884), left side 62 (value: 0) is a trough, 1/4 cycle is 10. According to the above, we determined right flank the trough is 82. According to the analysis result, carries on the binaryzation to Figure 4, the threshold value is 82, may extract iris’s long radius, as shown in Figure 5.
Figure 5 iris long radius
To Figure 1 image data, from about scans each picture element spot from top to bottom in order, (2) calculates each picture element according to the type and the center of circle distance.
(2)
And, dist is a distance, (x, y) is the flying spot coordinate figure, (Xpos, Ypos) is the iris center of circle coordinate figure. Retains is smaller than was equal to that the iris long radius or is bigger than was equal to the iris semi-minor axis the picture element, other suppose the picture element value are 0 (i.e. sign for black). The retention ring-like part namely for the iris image part which intercepts, as shown in Figure 6.
Figure 6 ring-like iris image part
In order to withdraw the iris image the characteristic value, establishes an eigen matrix array, X, Y value is consistent with the previous step rectangular array, uses for to deposit the corresponding characteristic value. These values are it are in sole possession, can carry on the only symbol to it regarding iris image’s in each picture element the value, therefore may use as the characteristic value. In what this article withdraws is each picture element two derivative takes its characteristic value, therefore may read in directly in this step it to the characteristic rectangle array.
Characteristic match
This article uses sea Ming Ju (Hamming Distance) to carry on the characteristic match. The sea bright distance to solve the error code problem which in the correspondence exists to invent at first. Simply speaking, it is refers to the similar length in two codes, corresponding position different code integer. For instance: 10101 and 00110, sea Ming Juwei 3. The type (3) is the formula which sea Ming Ju defines.
(3)
And Ai and Bi are treat the comparison the both sides code, is the “or else”, L is the code length.
Carries on when two iris image’s feature coding according to the position comparison, identical iris’s different time extraction condition code, its HD distributed peak value nearby 0.1; When the different iris’s condition code carries on compares to, HD distributed peak value nearby 0. 5. Here said when the distributed peak value is according to the position comparison, two section of feature coding corresponding phase with probability maximum value. Therefore, to the iris image eigen matrix array which already obtained, must first a stochastic choice section of L length code (binary system), namely the random selection code section reference. What here must pay attention, regarding treats the recognition two sections of codes, the reference must be as far as possible consistent. The L value may establish at will, but the L value is bigger, the match time is longer, the speed is fuller, the recognition precision is higher, the match accuracy is bigger; Otherwise, the L value is smaller, the match time are less, the speed is quicker, the recognition precision is lower, the match accuracy is smaller. In this article the L value supposes is 2048.
Result analysis
The accuracy is a most important performance index, generally indicated with the recognition rate, mainly by resists sentencing rate, the miscarriage of justice rate and so on leads to determine by mistake.
Resists sentencing rate FRR: Also said that resists reading rate wrongly or said the wrong match rate, did not indicate is authorized the human (legitimate user) not to acknowledge accurately (takes for assuming a false identity) degree. FRR is bigger, the system is more precise, the security is also higher, but the latitude is getting more and more low, cause more and more validated users by systematic error rejection. Otherwise is authorized the human to be easier to pass, the authorization also has not become easy to mix. FRR in fact is also the system acceptable important target.
Miscarriage of justice rate FAR: Also said that the wrong receive rate or said the wrong match rate, expressed the authorized person (assuming a false identity) had not been confirmed is authorized the human (effective individual) degree. The FAR value is smaller, showed that the authorized person more has not been unable to pass, the system is safer. But, will be authorized the human through will become even more difficult. If to has the strict request application domain safely, may move on very small FAR. Between FRR and FAR relations as shown in Figure 7.
Figure 7 resists sentencing rate and the miscarriage of justice rate between relations
The experimental result indicated that this article designs the system in the accuracy, the recognition speed has satisfied the practical request.
edward20071207110745.jpg
edward20071207110800.jpg
edward20071207110851.jpg
edward20071207110921.jpg
edward20071207111023.jpg
edward20071207111053.jpg
edward20071207111112.jpg
edward20071207111150.jpg
edward20071207111206.jpg
edward20071207111250.jpg
