Detection of non-ideal iris images based on meta-heuristic algorithms

Dissertation for receiving a Master's Degree in Mechatronics Engineering (M.Sc)

Abstract

Biometric technology, based on the unique characteristics of each person, automatically recognizes the identity of people. Researchers have been able to extract the iris tissue with high precision even in different conditions. As a result, our effort in this thesis is to present views and methods that improve the efficiency and accuracy of the identity recognition system. In the first approach, for the optimal zoning of the iris, it was necessary to determine the exact position of the center and the radius of the two circles to separate the iris pixels from other parts of the image, so for the optimal zoning of the iris using the ant colony algorithm, 4 optimization parameters were defined, and this solution finally led to the production of a new solution for iris detection. In the second approach, iris tissue features were extracted using two-dimensional violet and feature selection was done based on the artificial bee colony algorithm. The solutions presented to improve the number of features were associated with a significant reduction of features. In the third approach, the effect of different kernel types on classification accuracy with SVM neural network was investigated. Finally, the proposed methods performed better in comparison with other implemented methods. The results obtained on the images from the UBIRIS.v1 database, which includes 1877 photos of 241 people in JPEG format. Introduction

With the increase in the speed of information transmission and the emergence of electronic communications, information security plays an important role in protecting them and protecting people's privacy. In many of today's systems, information security may not be that complicated due to the confidentiality level of that system's data, but as the confidentiality level increases, the necessary security for information must be more complex. With the complexity of the necessary security mechanisms, perhaps other existing simple methods cannot provide the desired security. For this reason, the use of more important and more reliable parameters provides a higher level of security.

One of the most important options is the use of biometric components or unique body parameters, which is one of the most reliable choices, and one of the main factors in choosing these parameters is that they remain constant throughout human life. Biometric systems use the physiological or behavioral characteristics of each person to recognize their exact identity. In the human body, the eye is an external structure, but it is protected in a way, and it is an organ that rarely changes with the passage of time, and this feature makes this identification method more ideal than other methods.

Identity recognition through the iris as a reliable and powerful biometric technology has been able to attract a lot of attention. The complex texture of the iris and its unquestionable stability promise the use of iris-based identity recognition systems in various applications such as border control, court investigations, airports, security offices such as banks and government offices, and cryptography. The use of other visual features and facial features in addition to the iris can enable remote biometric identity recognition with appropriate accuracy, and the future of identity recognition with the help of the iris looks very bright, especially in military applications that require rapid identification of people in lively environments.

Properties in the iris due to random genetic changes are unique and experimental studies in large scales also confirm this theory among the studied population. When high security is considered, iris detection is the preferred method for identification because iris patterns are hardly affected by the environment and are hardly lost and can be widely used, so the need to implement meta-heuristic algorithms is necessary and this thesis can be a great help in this field.

1-2.. The purpose of the thesis

The use of meta-heuristic algorithms in identity recognition through the iris is considered as a new approach in this thesis and was introduced to achieve goals such as better and more accurate diagnosis in military, medical and security applications. This reliable and powerful biometric technology is able to solve the problems of organizations and industries such as the Ministry of Defense and the Armed Forces, airports, medical applications, electronic banking, computer system security, identity recognition and judicial applications.

Our goal is to provide a method for identifying people by extracting iris features by combining meta-heuristics and image processing methods.

To achieve this, the following steps are performed:

1. Choosing iris images 1

2. Pre-processing to fix image defects and errors 2

3. Iris localization

4. Iris image segmentation 3 means specifying the internal and external borders of the iris in the eye images

1- Image Selection

2- Preprocessing       

3- Iris segmentation

5. Postprocessing

6. Normalization means mapping the iris image to a rectangular strip with specific dimensions

7. Extracting feature 4 and selecting feature 5 means coding the features (extracting the feature vector from the normalized image) and using the codes containing the features extracted from the iris in order to recognize the identity (comparing the code of a person with other codes in the database in order to find the identity of the person among the stored information).

8. Classification 6: After the previous steps, the images are analyzed using quantitative methods so that each pixel is assigned to a specific class. characteristics of each individual recognizes the individual's identity automatically. Researchers exclusively by applying various methods could extract texture of iris with high accuracy even in different conditions. As a result, we in this thesis try to represent points of views and methods which improve the accuracy of identity recognition system. At the first trend for optimum partitioning of iris, there is a need for accurate location of center and radius of the two circles for isolation of iris pixels from the rest of the image's points. Therefore, to optimize iris segmentation by means of Ant Colony Algorithm, 4 optimization parameters were defined which resulted in producing a new solution to recognize iris. In the second trend, the method of characteristic extraction by means of two-dimensional violet based on the reduction of number in characteristic of artificial bees colony algorithm was suggested which the represented solutions to improve amount of this characteristic was represented and was along with reduction of these specifications and lowering amount of cost. In the third trend, the effect of kernel type in the accuracy of categorization with neuron network of SVM was investigated. At the end, the suggested methods in comparison with other methods represented better operation. The obtained results on the images from data station of UBIRIS.v1 included 1877 pictures from 241 people with JPEG format.