Presenting a method for identity recognition based on the feature of the fractal dimension

Computer Software Engineering Master's Thesis

Abstract:

Providing a method for identity recognition based on fractal dimension features

Introduction: Since a long time, there have been many techniques to identify people. Meanwhile, biometric-based identity recognition systems are more accurate and reliable. These systems should be designed in such a way that they can reliably identify people based on their physical and behavioral characteristics. Since relatively new methods such as finger veins have advantages such as high durability, user-friendliness and low risk in forgery compared to traditional biometric methods (such as fingerprints and iris), its use can provide better performance. Materials and methods: The proposed identification system consists of three main parts, which are: applying a mask to the image, extracting and matching the feature vector. The most important part in any biometric identity recognition system is the extraction of unique features, and in this research, fractals and subsequent calculation methods (such as the differential box counting method, differential box counting shift, relative differential box counting and multifractal) have been used for this purpose. Two databases, SDUMLA_HMT and FV_USM, which include images of index, ring, and index finger veins from both hands of people (male and female), have been used to apply the proposed method. All results have been prepared on MATLAB software version 2010.

Results: All performance parameters of the proposed method have been compared with existing methods in the field of identity recognition based on fractal and non-fractal models. The percentage of relative error rate on the databases of the mentioned images is equal to 0.07% and 0.1%, as well as the overall success rate of the system is equal to 99.85% and the FAR and FRR error rates are equal to 0.02% and 0.1%, respectively. Discussion: Our findings refer to the method of extracting features of biometric images using the concept of fractal dimension, which provides better results compared to existing methods. gives It should be noted that the proposed method has been improved with a completely new method that has not been observed in other researchers' studies.

Key words: biometrics, finger veins, people identification, fractal dimension, feature extraction.

In their daily life, people can recognize their friends and acquaintances according to their facial features, voice and even the way they walk. In fact, all people have special and unique characteristics that distinguish them from others. These characteristics and their study have led to the emergence of a branch of science called biometric science. Biometric science has a long history in authenticating people. In the era when the computer was not yet invented and the automatic tools developed today did not exist, biometric science was used with traditional and non-automatic methods. For many years, documents such as birth certificates and identification cards were used as documents to identify people. Following the expansion of the virtual world and electronic tools, the use of passwords and digital codes to do various things has found a special place. The rapid developments and widespread use of tools on the one hand and people's desire for the virtual world on the other hand have caused security to become highly important in various systems. The use of passwords, along with the advantages of its use, face challenges such as the possibility of being forgotten or revealed. Therefore, biometric science has opened a new window to a safe and secure world, where it authenticates people with fast and low-risk methods that can neither be stolen nor forgotten. The meaning of authentication is to confirm the correctness of data and information, which is generally done in different ways, which include:

Authentication based on documents, documents or tools that a person carries with him (token-based).

Authentication based on information that people are aware of (based on private knowledge).

Authentication of a person based on what he is (based on biometrics).

Biometric factors are classified into two general categories: behavioral factors and physical factors. The first category includes features such as the pattern of keyboard strokes, the sound pattern, the way of walking and so on. and in the second category features such as face scan, iris scan, heart rate pattern and so on.are checked The results indicate that physical factors have shown better efficiency than behavioral factors. The biometric factor must be such that it does not change under environmental conditions and with the passage of life. On the other hand, it should be usable by the general public. One of the important parameters in identity recognition systems is the efficiency of the biometric agent in terms of speed, cost and accuracy. It can be said that the biometric agent is suitable and more effective than other agents if the identification process can be done with higher speed and accuracy and less cost. In fact, a biometric system detects patterns by measuring biometric factors. The process of identity recognition includes the general steps of image acquisition, feature extraction, and matching and decision making. The most important part in this process is extracting effective and appropriate features, which is done by image processing operations and mathematical relationships. Using the features extracted from the patterns, a feature vector is generated, which is stored in a database for matching and decision making. Of course, there are several methods for feature extraction, one of which relies on calculating the fractal dimension of images. Since in this research, feature extraction has been done with a fractal-based strategy, a brief description of fractal and its dimension will be given below. All of them have a repetitive and chaotic structure. The study of these structures and the discovery of repeating patterns and their mathematical relationships have led to the emergence of fractal geometry. Fractals were first discovered and introduced by an English scientist named Mandelbrot. Essentially, a fractal consists of subsets that are similar in part to the whole. This feature is called self-similarity, the degree of self-similarity is different in different forms. Because of this feature, the fractal object looks the same from far and near. For example, if a piece of fern leaf is cut and magnified with a scale, it will finally show all the features and details of its original shape. The reason for this is the expansion of the details of the object in all dimensions and directions, which is referred to as objective or complete self-similarity. Volumes like cubes and cylinders are included in Euclidean geometry. These shapes follow special laws and mathematical relationships that cannot be used to describe fractal shapes. In other words, Euclidean geometry is unable to express the characteristics and examine the properties of fractals. Therefore, fractal geometry emerged to compensate for this deficiency. All volumes in Euclidean geometry have correct dimensions, for example, a line has a dimension of one, a plane has a dimension of two, and a cube has a dimension of three. This is because it is not possible to determine the correct dimension for the fractal object, but they have an incorrect and decimal dimension. As an example, Serpinski's triangle has a dimension equal to 1.58, which geometrically means that it is between a line and a plane (its degree of complexity is between a line and a plane). The meaning of dimension is to express the level of complexity and unevenness in an object. It is even possible to predict their behavior in the future by calculating the fractal dimension of a set of data. Such as the use of fractals in examining time series and forecasting the stock market. In general, in fractal geometry, any object that has the characteristics of self-similarity, decimal dimension, and micro-scale complexity is known as fractal. Several methods have been described for calculating the fractal dimension, the most famous and widely used of which can be called the differential box counting dimension. The methods of relative differential box counting dimension, box counting dimension by applying shift and . They are stated as corrective methods. The present research uses these methods to calculate the dimensions of images for identity recognition. 1-3- Objectives and structure of the thesis Identity recognition is one of the issues that are connected with people's lives and by using techniques, it has been tried to reduce the error and increase the confidence factor in their identification. Many biometric methods such as fingerprint, iris, face and. Among them, the finger vein is one of the physical features that has entered the field of identity recognition in recent years and has shown good performance. One of its important features is its high durability and no easy forgery. Therefore, finger vein patterns have been used in all stages of this research.