A privacy-preserving iris recognition method and system based on genetic algorithm
Through the privacy protection method based on genetic algorithm, the negative database and Heming distance are used to transform and protect iris data, the problem of irreversible iris data leakage is solved, and efficient privacy protection and identification performance is achieved.
Patent Information
- Application Number
- CN202210188169.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2022-02-28
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2042-02-28
AI Technical Summary
Existing iris recognition technology has flaws in protecting user privacy, especially once iris data is leaked, it is difficult to revoke, and traditional methods are difficult to effectively prevent reverse data recovery, resulting in high risk of privacy leakage.
Using a privacy protection method based on genetic algorithm, by converting the iris data submitted by users into a privacy-protected iris template, and using negative databases and Heming distances for identification, ensuring that the data is difficult to recover in reverse, combining roulette selection method and mutation operations , generate a high-fitting iris template.
It realizes iris recognition that effectively protects user privacy while ensuring data security. The generated iris template is difficult to reverse and has good recognition performance. It can effectively protect privacy while achieving accurate identity authentication.
Smart Images

Figure CN114550280B_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of privacy protection and security, and in particular relates to a privacy protection iris recognition method and system formed by improving a traditional iris recognition method using a genetic algorithm. Background Art
[0002] With the rapid development of Internet technology, our lives are moving towards intelligence and digitalization. The widespread application of emerging Internet technologies such as big data has provided us with various conveniences, but it has also brought some unavoidable problems, the most important of which is the leakage of user privacy. In recent years, the frequent leakage of user privacy data at home and abroad has further increased people's attention to their own privacy data. Once a large amount of user data is leaked, it will not only cause huge economic losses to society or enterprises, but also cause great harm to the physical and mental health of users whose privacy has been leaked. Therefore, how to conduct various scientific research under the premise of ensuring data security has become a hot topic for scholars to study and explore, and it is also a very practical research direction.
[0003] Identity authentication based on biometric information has become a research hotspot in recent years. Among them, iris recognition, as a very important biometric information recognition method, has many good characteristics in terms of identity characteristics. However, human biometric data often faces serious privacy leakage problems, especially since most of the biometric information such as iris remains unchanged throughout a person's life. Once this data is leaked, it will lead to permanent loss of data and cannot be undone. Therefore, it is very important to protect data security during the iris recognition process.
[0004] Evolutionary algorithm is an optimization algorithm inspired by the biological evolution mechanism. The algorithm starts from a random initial population, and through a series of operations that imitate genetic inheritance and mutation mechanisms, it finally searches for the optimal solution or reaches the maximum number of iterations to obtain a solution that satisfies the conditions. In a manner similar to natural selection and sexual reproduction, it generates a next generation population with higher environmental adaptability while retaining the original excellent genes. Among them, the genetic algorithm, one of the classic algorithms in the evolutionary algorithm, has been successfully applied to different fields such as job scheduling, automatic control, artificial intelligence and computer science, and social economy, and has solved many problems. Therefore, it is of great significance to explore the application of genetic algorithms in privacy-preserving iris recognition solutions.
[0005] Negative database is the main model of negative information representation. It protects private data by storing data in the complement of the original data set, thereby achieving privacy protection. It is very difficult to restore the original data from the negative database. Therefore, even if the attacker obtains the data in the negative data, it is difficult to obtain the real data in the original database. Therefore, the negative database can be effectively used to protect user privacy. At present, negative databases have been widely used in negative authentication of password information, negative investigation, privacy protection, data mining, etc. Therefore, it is very valuable to explore new methods of negative database in privacy-preserving iris recognition. Summary of the invention
[0006] In order to overcome the defects of the above-mentioned background technology, the present invention provides a privacy-preserving iris recognition method based on genetic algorithm, in order to explore a new solution for privacy-preserving iris recognition.
[0007] In order to solve the above technical problems, the technical solution adopted by the present invention is:
[0008] A privacy-preserving iris recognition method based on a genetic algorithm, characterized by comprising:
[0009] In the registration phase, the iris data t submitted by the user is obtained, the data legitimacy is checked, the original iris data t submitted by the user is converted into a privacy-protected iris template t′, and added to the iris template database, and a registration success message is returned;
[0010] In the recognition stage, the real-time collected iris data d submitted by the user is obtained, and the Hamming distance between d and the iris template in the iris template database is calculated. When the Hamming distance is less than or equal to the threshold C, the recognition is successful. If the Hamming distance between all templates in the template database and the iris data d is greater than C, the recognition fails.
[0011] Preferably, the registration phase includes:
[0012] Step 11, initialize the population with size population_size;
[0013] Step 12: For each individual in the population, the crossover probability P c Make a crossover;
[0014] Step 13: For each individual in the population, according to the mutation probability P v Perform mutation operations;
[0015] Step 14, calculate the fitness of each individual in the population according to the evaluation function f(x)=unmatch(x,NDB);
[0016] Step 15, calculate the fitness of all individuals according to step 14, and then sort all individuals from high to low according to fitness; then select individuals of the next generation population according to the roulette wheel selection method;
[0017] Step 16, determine whether the maximum number of iterations has been reached. If so, select the individual with the largest fitness in the population as the iris template t′ that has been processed with privacy protection, and add it to the iris template database; if not, return to step 2.
[0018] Preferably, step 11 includes:
[0019] First, set the population size population_size, the value range of population size population_size is 50-200; then determine the chromosome length m, the chromosome length m is 1536 bits; select any position in the iris string to be converted that accounts for the proportion of the total string length between [0.05,0.2] and take the inversion to obtain a binary string with a distance from the original iris data between [0.05,0.2]; add the 99 initialized individual iris strings obtained by the above method and the original iris data, a total of 100 initial test individuals, to the initial population.
[0020] Preferably, the crossover probability P in step 12 is c It is artificial experience data and is set between 0.9-0.97 based on artificial experience.
[0021] Preferably, the mutation probability P in step 13 is v It is artificial experience data, which is set between 0.1 and 0.001 according to artificial experience. First, according to the mutation probability P v Decide whether to perform mutation on the chromosome. For the individual to be mutated, randomly select a position on the chromosome, invert '0' and '1' to obtain a new individual, and add the newly generated individual to the population.
[0022] Among them, according to the mutation probability P v The method for determining whether to perform a mutation operation on the chromosome includes: for each individual, generating a random number random to control whether to perform a mutation, and judging whether the random number random is less than the mutation probability P v , if yes, then mutation operation is performed, if no, then no mutation operation is performed.
[0023] Preferably, the evaluation function in step 14 is f(x)=unmatch(x,NDB), and the number of mismatches between the chromosome code string of each individual x in the population and the records in the negative database NDB is calculated, and this is used as the value of fitness; the value of individual fitness is proportional to the number of mismatches between the individual chromosome and the records in the negative database; the specific steps of calculating fitness using the fitness function are as follows:
[0024] First, the K-hidden negative data generation algorithm is used to convert a random string with the same length as the original iris data string into the negative database NDB; the main parameters of the K-hidden negative database generation algorithm are set as follows: the parameter K=3 that controls the number of definite bits in the negative database, the binary string length m=1536 used for conversion, the parameter r=15 that controls the size of the negative database, and the parameters p1=0.752, p2=0.226, and p3=0.02 that control the generation probability of K different types of records. According to the parameters, the negative database generated by the K-hidden algorithm contains a total of m×r=23040 records, each of which contains 3 definite bits, and the other positions are uncertain bits, which can take either '0' or '1'; the use of random strings to generate the negative database is to ensure that the scheme has sufficient security, and the negative database only stores the definite bits of several character strings;
[0025] Next, the number of mismatches between the chromosome code string x of each individual in the population and the records in the negative database is calculated as unmatch(x,NDB), which is used as the value of fitness. The value of individual fitness is proportional to the number of mismatches between the individual chromosome and the negative database.
[0026] Preferably, the method of selecting individuals of the next generation population according to the roulette wheel selection method in step 15 includes:
[0027] Step 151, calculate the fitness f(x i );
[0028] Step 152, calculate the probability P(x) of each individual being inherited to the next generation according to the following formula i ), where n is the total number of individuals before selection;
[0029]
[0030] Step 153, calculate the cumulative probability q of each individual according to the following i ;
[0031]
[0032] Step 154, generating a uniformly distributed pseudo-random number r in [0,1];
[0033] Step 155, if r < q[1], then select individual 1; otherwise, select individual k that satisfies q[k - 1] < r < q[k].
[0034] Step 156, repeat Step 154 and Step 155 for population_size times, so as to select population_size individuals to enter the next generation.
[0035] Preferably, the recognition stage includes: obtaining the iris data d collected in real time submitted by the user for recognition. The server calculates the Hamming distance between the iris data d and the iris templates in the iris template database. If the Hamming distance is less than or equal to the threshold C, then return recognition success. If the distances between all templates in the template database and the iris data d are greater than the threshold C, then return recognition failure.
[0036] Preferably, the method for the server to calculate the Hamming distance between the iris data d and the iris templates in the iris template database includes:
[0037]
[0038] where HD represents the Hamming distance, n represents the string length of the iris data, i represents the i-th position. Therefore, the subscripted d and t′ respectively represent the i-th positions of the iris string d and the template string t′. The symbol represents the exclusive OR operation, and the calculation method is that if the corresponding positions of d and t′ are the same, the result is "0"; if they are different, the result is "1".
[0039] The present invention also provides a privacy - protected iris recognition system based on the genetic algorithm using the above - mentioned method, including an iris registration module and an iris recognition module;
[0040] The iris registration module is used to obtain the iris data t submitted by the user, check the data legality, convert the original iris data t submitted by the user into a privacy - protected iris template t′, and add it to the iris template database, and return the registration success information;
[0041] The iris recognition module is used to obtain the iris data d collected in real time submitted by the user, calculate the Hamming distance between d and the iris templates in the iris template database. When the Hamming distance is less than or equal to the threshold C, the recognition is successful. If the Hamming distances between all templates in the template database and the iris data d are greater than C, then the recognition fails.
[0042] The beneficial effects of the present invention are: (1) a genetic algorithm is used to perform iris privacy protection processing. Since there is a great deal of randomness in the evolution process of the genetic algorithm, the iris template generated by the transformation is difficult to recover; (2) a new information security technology such as a negative database is introduced to design the evaluation function in the genetic algorithm, making the transformation process more difficult to reverse; (3) the privacy protection iris recognition scheme proposed by the present invention has good recognition performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0043] Figure 1 This is the system architecture diagram of iris registration and recognition of the present invention,
[0044] Figure 2 This is a flowchart of the implementation of the privacy protection processing of the original iris data in the present invention. DETAILED DESCRIPTION
[0045] The present invention will be further described below in conjunction with the accompanying drawings and embodiments.
[0046] In order to explore a new solution for privacy-preserving iris recognition, the present invention proposes a privacy-preserving iris recognition method based on a genetic algorithm.
[0047] A privacy-preserving iris recognition method based on a genetic algorithm, characterized by comprising:
[0048] In the registration phase, the iris data t submitted by the user is obtained, the data legitimacy is checked, the original iris data t submitted by the user is converted into a privacy-protected iris template t′, and added to the iris template database, and a registration success message is returned;
[0049] In the recognition stage, the real-time collected iris data d submitted by the user is obtained, and the Hamming distance between d and the iris template in the iris template database is calculated. When the Hamming distance is less than or equal to the threshold C, the recognition is successful. If the Hamming distance between all templates in the template database and the iris data d is greater than C, the recognition fails.
[0050] Preferably, the registration phase includes:
[0051] Step 11, initialize the population with size population_size;
[0052] Step 12: For each individual in the population, the crossover probability P c Make a crossover;
[0053] Step 13: For each individual in the population, according to the mutation probability P v Perform mutation operations;
[0054] Step 14, calculate the fitness of each individual in the population according to the evaluation function f(x)=unmatch(x,NDB);
[0055] Step 15, calculate the fitness of all individuals according to step 14, and then sort all individuals from high to low according to fitness; then select individuals of the next generation population according to the roulette wheel selection method;
[0056] Step 16, determine whether the maximum number of iterations has been reached. If so, select the individual with the largest fitness in the population as the iris template t′ that has been processed with privacy protection, and add it to the iris template database; if not, return to step 2.
[0057] Preferably, step 11 includes:
[0058] First, set the population size population_size, the value range of population size population_size is 50-200; then determine the chromosome length m, the chromosome length m is 1536 bits; select any position in the iris string to be converted that accounts for the proportion of the total string length between [0.05,0.2] and take the inversion to obtain a binary string with a distance from the original iris data between [0.05,0.2]; add the 99 initialized individual iris strings obtained by the above method and the original iris data, a total of 100 initial test individuals, to the initial population.
[0059] Preferably, the crossover probability P in step 12 is c It is artificial experience data and is set between 0.9-0.97 based on artificial experience.
[0060] Preferably, the mutation probability P in step 13 is v It is artificial experience data, which is set between 0.1 and 0.001 according to artificial experience. First, according to the mutation probability P v Decide whether to perform mutation on the chromosome. For the individual to be mutated, randomly select a position on the chromosome, invert '0' and '1' to obtain a new individual, and add the newly generated individual to the population.
[0061] Among them, according to the mutation probability P v The method for determining whether to perform a mutation operation on the chromosome includes: for each individual, generating a random number random to control whether to perform a mutation, and judging whether the random number random is less than the mutation probability P v , if yes, then mutation operation is performed, if no, then no mutation operation is performed.
[0062] Preferably, the evaluation function in step 14 is f(x)=unmatch(x,NDB), and the number of mismatches between the chromosome code string of each individual x in the population and the records in the negative database NDB is calculated, and this is used as the value of fitness; the value of individual fitness is proportional to the number of mismatches between the individual chromosome and the records in the negative database; the specific steps of calculating fitness using the fitness function are as follows:
[0063] First, the K-hidden negative data generation algorithm is used to convert a random string with the same length as the original iris data string into the negative database NDB; the main parameters of the K-hidden negative database generation algorithm are set as follows: the parameter K=3 that controls the number of definite bits in the negative database, the binary string length m=1536 used for conversion, the parameter r=15 that controls the size of the negative database, and the parameters p1=0.752, p2=0.226, and p3=0.02 that control the generation probability of K different types of records. According to the parameters, the negative database generated by the K-hidden algorithm contains a total of m×r=23040 records, each of which contains 3 definite bits, and the other positions are uncertain bits, which can take either '0' or '1'; the use of random strings to generate the negative database is to ensure that the scheme has sufficient security, and the negative database only stores the definite bits of several character strings;
[0064] Next, the number of mismatches between the chromosome code string x of each individual in the population and the records in the negative database is calculated as unmatch(x,NDB), which is used as the value of fitness. The value of individual fitness is proportional to the number of mismatches between the individual chromosome and the negative database.
[0065] Preferably, the method of selecting individuals of the next generation population according to the roulette wheel selection method in step 15 includes:
[0066] Step 151, calculate the fitness f(x i );
[0067] Step 152, calculate the probability P(x) of each individual being inherited to the next generation according to the following formula i ), where n is the total number of individuals before selection;
[0068]
[0069] Step 153, calculate the cumulative probability q of each individual according to the following i ;
[0070]
[0071] Step 154, generating a uniformly distributed pseudo-random number r in [0,1];
[0072] Step 155: If r < q[1], then select individual 1; otherwise, select individual k that satisfies q[k - 1] < r < q[k].
[0073] Step 156: Repeat Step 154 and Step 155 for population_size times, so as to select population_size individuals to enter the next generation.
[0074] Preferably, the recognition stage includes: obtaining the iris data d collected in real time submitted by the user for recognition, the server calculating the Hamming distance between the iris data d and the iris templates in the iris template database. If the Hamming distance is less than or equal to the threshold C, then return recognition success; if the distances between all templates in the template database and the iris data d are greater than the threshold C, then return recognition failure.
[0075] Preferably, the method for the server to calculate the Hamming distance between the iris data d and the iris templates in the iris template database includes:
[0076]
[0077] Wherein, HD represents the Hamming distance, n represents the string length of the iris data, i represents the i-th position. Therefore, the subscripted d and t' respectively represent the i-th position of the iris string d and the template string t'. The symbol represents the exclusive OR operation, and the calculation method is that if the corresponding positions of d and t' are the same, the result is "0"; if they are different, the result is "1".
[0078] A privacy - protected iris recognition system based on the genetic algorithm using the above - mentioned method, including an iris registration module and an iris recognition module;
[0079] The iris registration module is used to obtain the iris data t submitted by the user, check the data legality, convert the original iris data t submitted by the user into a privacy - protected iris template t', and add it to the iris template database, and return the registration success information;
[0080] The iris recognition module is used to obtain the iris data d collected in real time submitted by the user, calculate the Hamming distance between d and the iris templates in the iris template database. When the Hamming distance is less than or equal to the threshold C, the recognition is successful; if the Hamming distances between all templates in the template database and the iris data d are greater than C, then the recognition fails.
[0081] The following are specific embodiments:
[0082] The technical solutions of the registration stage, recognition stage, and performance evaluation of the method in this embodiment are as follows:
[0083] During the registration phase, the user submits an iris data t to the server to request registration. The server first checks the data legitimacy. The server uses the privacy-preserving iris template generation algorithm designed based on the genetic algorithm to convert the original iris data submitted by the user into a privacy-preserving iris template t′, and adds it to the iris template database, returning a successful registration.
[0084] Specifically, the registration phase includes the following steps:
[0085] Step 11: Initialize a population of size population_size
[0086] First, the population size population_size is set according to the application scenario. The larger the population size, the less likely it is that an individual optimal solution will dominate the evolution direction of the overall solution. The smaller the population size, the slower the optimal solution is found, and the more likely it is to fall into a local optimal solution. The size of the population is related to the complexity of individual genes. The specific size can be estimated by continuous trial and error, which is generally around 50 to 200. Therefore, in this solution, the population size population_size is set to 100. Next, the chromosome length m is determined. Since this method is proposed to protect iris data, and the iris data in the dataset CASIA-IrisV3-Interval is converted into a binary string with a length of 1536 bits after processing and preprocessing, the chromosome length m is determined to be 1536 bits. In the iris string to be converted, select any position that accounts for the total string length between [0.05, 0.2] and take the inversion (this ratio can be adjusted according to the requirements of security and performance), and then obtain a binary string with a distance between [0.05, 0.2] from the original iris data. Specifically, any 76 to 460 bits of the 1536-bit original iris data binary are randomly selected and flipped, and the original '0' is flipped to '1', and the original '1' is flipped to '0'.
[0087] After the above method is used to process, 99 initialized individual iris strings are obtained, and a total of 100 initial test individuals are added to the initial population together with the original iris data, thus ensuring that the initialized population is highly similar to the original iris data.
[0088] Step 12: For each individual in the population, c Cross
[0089] There are two ways to set the crossover probability: the first is to take a fixed value. Generally speaking, the crossover probability is between 0.9 and 0.97. The second is to set it adaptively, according to the fitness value of the crossover or mutation individual and the average fitness value of the current generation. The individuals of each generation are different. The relevant formula can be obtained by looking up the data. In this scheme, the crossover probability P is initialized. cIt is set to 0.95, and the crossover probability is used to determine whether to perform a crossover operation on a specific individual. If a crossover operation is required, a crossover position is randomly selected on the chromosome, and all gene segments at and after that position are exchanged with the alleles of the chromosome it is paired with to generate two new individuals and add them to the population.
[0090] Step 13: For each individual in the population, according to the mutation probability P v Perform mutation operation.
[0091] Like the crossover probability, there are two ways to set the mutation probability: the first is to take a fixed value, usually the mutation probability is between 0.1 and 0.001; the second is to take an adaptive value, which is calculated according to the fitness value of the crossover or mutation individual and the average fitness value of the current generation. The individuals of each generation are different. In this scheme, the mutation probability P is initialized. v Set to 0.01. First, decide whether to mutate the chromosome based on the mutation probability. Since the encoding method used in this problem is binary encoding, the individual that meets the crossover condition randomly selects a position on the chromosome, inverts '0' and '1', and then generates a new individual, which is added to the population.
[0092] Step 14: Calculate the fitness of each individual according to the evaluation function.
[0093] The design of the evaluation function is a key point of the present invention. The specific design method is as follows: first, a K-hidden negative data generation algorithm is used to convert a random string with the same length as the original iris data string into a negative database (NDBs). The main parameters of the K-hidden negative database generation algorithm are set as follows (under this parameter setting, the generated negative database can meet higher security requirements, so the following parameters are used): the parameter K=3 for controlling the number of definite bits in the negative database, the binary string length m=1536 for conversion, the parameter r=15 for controlling the scale of the negative database, and the parameters p1=0.752, p2=0.226, and p3=0.02 for controlling the generation probability of K different types of records. According to the parameters, the negative database generated by the K-hidden algorithm contains a total of m×r=23040 records, each of which contains 3 definite bits, and the other positions are uncertain bits, and the uncertain bits can take either '0' or '1'. The negative database is generated by using random strings to ensure that the scheme has sufficient security. The negative database only stores the exact locations of several character strings. Next, the number of mismatches between the chromosome encoding string of each individual in the population and the records in the negative database is calculated, and this is used as the value of fitness. The higher the number of mismatches between the individual's chromosome and the records in the negative database, the higher the corresponding fitness value. This is used to judge the adaptation of individuals in the population, and then provide a basis for the survival of the fittest selection mechanism of the evolutionary algorithm.
[0094] Step 15: Use appropriate selection mechanisms to select individuals with outstanding performance to enter the next generation.
[0095] First, the fitness of all individuals is calculated according to the previous step, and then all individuals are sorted from high to low according to their fitness. Next, the selection is made according to the "roulette selection method". The probability of each individual being selected is proportional to its fitness. Individuals with higher fitness have a higher probability of being selected into the next generation population.
[0096] Step 16: Determine whether the iteration is completed.
[0097] When initializing the parameters, the maximum number of iterations N is set to 20. If the maximum number of iterations is reached, the individual with the largest fitness in the population is selected as the iris template t′ that has been processed with privacy protection, and it is added to the iris template database; otherwise, return to step 2 and continue the operations of crossover, mutation, selection, etc.
[0098] In the recognition phase, the user submits a real-time collected iris data d to request recognition, and the server calculates the Hamming distance between d and the iris template in the iris template database. When the distance is less than or equal to the threshold C, the recognition is successful. If the distance between all templates in the template database and d is greater than C, the recognition fails. The experimental results show that the present invention can perform iris recognition under the premise of effectively protecting the security of user iris data and achieve good recognition effect.
[0099] The identification phase includes the following steps:
[0100] Step 21: The user submits a piece of iris data d to the server to request iris recognition.
[0101] The server calculates the Hamming distance between the iris data d and each iris template t′ in the database. The Hamming distance calculation formula is:
[0102]
[0103] The symbol represents the XOR operation. The specific algorithm for the XOR operation is: for each corresponding position of the iris string d and the template string t′, if the values at the corresponding positions are the same, that is, both are '1' or both are '0', the XOR result is '0'; if the values at the corresponding positions are different, the XOR result is '1'. Sum the XOR results of all corresponding positions and divide them by the string length to get the Hamming distance value HD.
[0104] Step 22: Return the recognition result.
[0105] The system pre-sets a threshold C of Hamming distance based on the trade-offs of security and user experience. If the calculated Hamming distance HD is less than or equal to the set threshold C, the recognition is successful and the user is allowed to access the server resources. If the distance between d and all templates in the template database is greater than C, the recognition is failed.
[0106] After the recognition stage, this embodiment also includes a performance evaluation scheme. Since the purpose of the privacy-preserving iris recognition method based on genetic algorithm is to protect the template data in the user's template database, this scheme does not involve image preprocessing, iris feature extraction and other processing processes in the iris recognition process, but directly uses the open source iris data preprocessing system OSIRIS to convert the iris image into a binary iris string for performance evaluation to verify the effectiveness of the privacy-preserving iris recognition method designed by the present invention. The data set used for the performance evaluation of this method is the data in the commonly used iris database CASIA-IrisV3-Interval, and the data in the data set includes iris data of 249 people. In the specific implementation process, 1332 iris data of the left eyes of all people in the data set are used, which are divided into 198 classes in total. Each class includes several iris data of the same person, and the iris data is iris data with a string length of 1536 bits.
[0107] In order to verify the effectiveness of the privacy protection model proposed in the present invention, two performance indicators, False Acceptance Rate (FAR) and Genuine Acceptance Rate (GAR), are introduced to quantitatively evaluate the iris recognition performance of the present invention. Under a specific matching threshold, FAR indicates the percentage of times that iris data from different categories are mistakenly identified as coming from the same category in the total number of tests; GAR indicates the percentage of times that iris data are identified as the correct category in the total number of tests. The calculation formulas for FAR and GAR are as follows:
[0108]
[0109]
[0110] The present invention designs a privacy-preserving iris recognition method based on genetic algorithm. Experimental results show that the privacy-preserving iris recognition method based on genetic algorithm designed by us can achieve good recognition performance under the premise of protecting user privacy.
[0111] In summary, the present invention has the following advantages: (1) Genetic algorithm is used to perform iris privacy protection processing. Since there is a lot of randomness in the evolution process of genetic algorithm, the transformed iris template is difficult to recover; (2) The negative database, a new information security technology, is introduced to design the evaluation function in the genetic algorithm, making the transformation process more difficult to reverse; (3) The privacy protection iris recognition scheme proposed in the present invention has good recognition performance.
[0112] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all these improvements and changes should fall within the scope of protection of the appended claims of the present invention.
Claims
1. A privacy-preserving iris recognition method based on genetic algorithm, characterized in that: include: In the registration phase, the iris data t submitted by the user is obtained, the data legitimacy is checked, the original iris data t submitted by the user is converted into a privacy-protected iris template t′, and added to the iris template database, and a registration success message is returned; In the recognition stage, the real-time collected iris data d submitted by the user is obtained, and the Hamming distance between d and the iris template in the iris template database is calculated. When the Hamming distance is less than or equal to the threshold C, the recognition is successful. If the Hamming distance between all templates in the template database and the iris data d is greater than C, the recognition fails. The registration phase includes: Step 11, initialize the population with size population_size; Step 12: For each individual in the population, the crossover probability P c Make a crossover; Step 13: For each individual in the population, according to the mutation probability P v Perform mutation operations; Step 14, calculate the fitness of each individual in the population according to the evaluation function f(x)=unmatch(x,NDB); Step 15, calculating the fitness of all individuals according to step 14, and then sorting all individuals from high to low according to fitness; and then selecting individuals of the next generation population according to the roulette wheel selection method; Step 16, determining whether the maximum number of iterations has been reached, if so, selecting the individual with the largest fitness in the population as the iris template t′ that has been processed with privacy protection, and adding it to the iris template database; if not, returning to step 12; The evaluation function in step 14 is f(x)=unmatch(x,NDB), which calculates the number of mismatches between the chromosome code string of each individual x in the population and the records in the negative database NDB, and uses this as the value of fitness; the value of individual fitness is proportional to the number of mismatches between the individual chromosome and the records in the negative database; the specific steps of calculating fitness using the fitness function are as follows: First, a K-hidden negative data generation algorithm is used to convert a random string with the same length as the original iris data string into a negative database NDB; the main parameters of the K-hidden negative database generation algorithm are set as follows: the parameter K=3 that controls the number of definite bits in the negative database, the binary string length m=1536 used for conversion, the parameter r=15 that controls the size of the negative database, and the parameters ρ1=0.752, ρ2=0.226, and p3=0.02 that control the generation probability of K different types of records. According to the parameters, the negative database generated by the K-hidden algorithm contains a total of m×r=23040 records, each of which contains 3 definite bits, and the other positions are uncertain bits, which can take either '0' or '1'; the use of random strings to generate the negative database is to ensure that the scheme has sufficient security, and the negative database only stores the definite bits of several character strings; Next, the number of mismatches between the chromosome code string x of each individual in the population and the records in the negative database is calculated as unmatch(x,NDB), which is used as the value of fitness. The value of individual fitness is proportional to the number of mismatches between the individual chromosome and the negative database.
2. The privacy-preserving iris recognition method based on genetic algorithm according to claim 1, characterized in that: The step 11 comprises: First, set the population size population_size, where the value range of the population size population_size is 50 - 200; then determine the chromosome length m, and the chromosome has a length m of 1536 bits; take the bitwise inversion at any position in the iris string to be transformed where the proportion of the total string length is between [0.05, 0.2], and obtain a binary string with a Hamming distance from the original iris data between [0.05, 0.2]; add the 99 initialized individual iris strings obtained through the above - mentioned method and the original iris data, a total of 100 initial individuals, to the initial population.
3. The privacy-preserving iris recognition method based on genetic algorithm according to claim 1, characterized in that: The crossover probability P in step 12 c It is artificial experience data and is set between 0.9-0.97 based on artificial experience.
4. The privacy-preserving iris recognition method based on genetic algorithm according to claim 1, characterized in that: The mutation probability P in step 13 v It is artificial experience data, which is set between 0.1 and 0.001 according to artificial experience. First, according to the mutation probability P v Decide whether to perform mutation on the chromosome. For the individual to be mutated, randomly select a position on the chromosome, invert '0' and '1' to obtain a new individual, and add the newly generated individual to the population. Among them, according to the mutation probability P v The method for determining whether the chromosome is to be mutated includes: for each individual, generating a random number random to control whether to mutate, and determining whether the random number random is less than the mutation probability P v , if yes, then mutation operation is performed, if no, then no mutation operation is performed.
5. The privacy-preserving iris recognition method based on genetic algorithm according to claim 1, characterized in that: The method of selecting individuals for the next - generation population in step 15 according to the roulette - wheel selection method includes: Step 151, calculate the fitness f(x i ); Step 152, calculate the probability P(x) of each individual being inherited to the next generation according to the following formula i ), where n is the total number of individuals before selection; Step 153, calculate the cumulative probability q of each individual according to the following i ; Step 154, generate a uniformly - distributed pseudo - random number r in the range [0, 1]; Step 155, if r < q[1], then select individual 1, otherwise select individual k that satisfies q[k - 1] < r < q[k]; Step 156, repeat step 154 and step 155, and the number of repetitions is population_size times, so as to select population_size individuals to enter the next generation; 6. The privacy-preserving iris recognition method based on genetic algorithm according to claim 1, characterized in that: The recognition stage includes: obtaining the iris data d submitted by the user for real - time acquisition for recognition, and the server calculates the Hamming distance between the iris data d and the iris templates in the iris template database. If the Hamming distance is less than or equal to the threshold C, then return recognition success. If the distances between all templates in the template database and the iris data d are greater than the threshold C, then return recognition failure.
7. The privacy-preserving iris recognition method based on genetic algorithm according to claim 1, characterized in that: The method for the server to calculate the Hamming distance between the iris data d and the iris templates in the iris template database includes: Where HD represents the Hamming distance, n represents the length of the iris data string, and i represents the i-th position. Therefore, the subscripts d and t′ represent the i-th position of the iris string d and the template string t′, respectively. It represents XOR operation. The calculation method is that if the corresponding positions of d and t′ are the same, the result is "0", and if they are different, the result is "1".
8. A privacy-preserving iris recognition system based on a genetic algorithm using the method according to any one of claims 1 to 7, characterized in that: Including an iris registration module and an iris recognition module; The iris registration module is used to obtain the iris data t submitted by the user, check the data legality, transform the original iris data t submitted by the user into a privacy - protected iris template t′, and add it to the iris template database, and return the registration success information; The iris recognition module is used to obtain the iris data d submitted by the user for real - time acquisition, calculate the Hamming distance between d and the iris templates in the iris template database. When the Hamming distance is less than or equal to the threshold C, the recognition is successful. If the Hamming distances between all templates in the template database and the iris data d are greater than C, then the recognition fails.
