Розпізнавання облич методами машинного навчання
| dc.contributor.advisor | Іванько, Катерина Олегівна | |
| dc.contributor.author | Бондарев, Вʼячеслав Ростиславович | |
| dc.date.accessioned | 2023-08-29T12:59:24Z | |
| dc.date.available | 2023-08-29T12:59:24Z | |
| dc.date.issued | 2021 | |
| dc.description.abstract | Дипломна робота присвячена алгоритмам машинного навчання в задачі ідентифікації особи за зображенням обличчя. Загальний обсяг роботи складає 75 сторінку, кількість ілюстрацій – 41, таблиць – 1, джерел за переліком посилань – 32. Метою роботи є розробити алгоритм, що буде включати обробку зображень, виділення ознак особи за зображенням і навчання класифікаторів для рішення задачі ідентифікації особи. У першому розділі проведено аналіз існуючих підходів до задачі біомедичної ідентифікації особи за зображенням обличчя і основних методів виділення ознак осіб за зображенням обличчя та обрано найкращий метод. У другому розділі проведено аналіз існуючих методів обробки зображень і алгоритмів цих методів. У третьому розділі проведено аналіз існуючих сценаріїв та задач машинного навчання та обрано найкращий сценарій навчання для задачі ідентифікації особи за зображенням обличчя. У четвертому розділі розроблено алгоритм цифрової обробки зображень і виділення ознак особи за методом головних компонент в середовищі MATLAB. У п’ятому розділі розроблено алгоритм класифікації особи за зображенням обличчя, проведено аналіз впливу різних факторів на точність класифікації. У висновках на підставі проведеного аналізу даних і розробленого алгоритму, наведено головні результати роботи та сформульовано рекомендації щодо їх практичного застосування. | uk |
| dc.description.abstractother | Thesis is devoted to machine learning algorithms in the problem of person identification by face image. The total volume of the work is 75 pages, the number of illustrations – 41, tables – 1, sources according to the list of references – 32. The aim of the work is to develop an algorithm that will include image processing, selection of facial features by image and training of classifiers to solve the problem of facial identification. In the first section the analysis of the existing approaches to the problem of biomedical identification of the person by the photo and the basic methods of allocation of signs of persons is carried out and the best method is chosen. The second section analyzes the existing methods of image processing and algorithms of these methods. The third section analyzes the existing scenarios and tasks of machine learning and selects the best learning scenario for the task of facial identification. In the fourth section, an algorithm for digital image processing and selection of facial features by the method of principal components in the MATLAB environment is developed. The fifth section develops an algorithm for classifying a person by facial image, analyzes the impact of various factors on the accuracy of classification. In the conclusions, based on the conducted analysis of data and the developed algorithm, the main results of work are resulted and recommendations concerning their practical application are formulated. Interests on the field of identification and authentication of the individual in the modern world is gaining more and more practical need: from systems of protection and verification of credit cards to forensic examination and identification of criminals. There are many approaches to the selection of facial features by the image, as well as to solve the problem of identification. While there are many different approaches to the problem of identifying personality traits, the most common solution to the problem of identification is the use of machine learning tools. Machine learning is a large unit of artificial intelligence that studies methods for constructing algorithms capable of learning. Machine learning is at the intersection of mathematical statistics, optimization methods and discrete mathematics, but also has its own specifics related to the problems of computational efficiency and retraining. In general, the classification algorithm consists of the following steps: pre-processing of images, selection of facial features by the image and training classification. The use of machine learning approach to the problem of biomedical identification of a person by facial image requires special attention to the input of images. Under the input of the image should be understood the procedure for converting the original image to a form suitable for a computer system. The use of these procedures is necessary to discard unimportant information to reduce the size of the input data array and highlight the most important features of the facial image for training of classifiers. In other words, digital processing of the array of input images is necessary to increase the speed and quality of further machine learning. In this problem, the most appropriate is to use image compression, convert it to grayscale and processing the brightness histogram. All information compression methods are based on the assumption that the data set always contains unimportant elements. Compression is achieved by finding and encoding redundant elements. Eliminating the visual redundancy of images is the main reserve for reducing the size of the input data array. To optimize the image size, it is necessary, on the one hand, not to use unimportant information, and on the other - to prevent excessive loss of image quality. The use of grayscale images can significantly reduce the size of the input data, because it allows you to replace the three brightness matrices of the RGB system with one matrix of white brightness. There are several basic conversion methods based on different weighing of color channels. Image processing may degrade the contrast of the image, which is not acceptable for the task of identifying a person by face image. The brightness of the image in many cases is determined by the contrast, which represents the difference between the maximum and minimum levels of halftones in some area of the image. By increasing the contrast, you can improve the visibility of special image structures or reduce the distortion that occurred when obtaining the image. Improving the contrast of individual parts of the face is a key factor in improving the accuracy of the problem of biomedical identification of a person by facial image. An image brightness histogram is a graph that shows the relative pixel frequency of different brightness values in an image. Histogram processing is the process of changing an image by changing its histogram. Typically, histogram processing involves normalization, which makes the histogram of the image as flat as possible. There are several basic methods for distinguishing facial features from an image, but the simplest and most common are the principal components method. The principal components method is a universal method that analyzes a data table in which dependencies are described by several interrelated quantitatively dependent variables. In the problem of face recognition, it is used mainly to represent the image of a person by a vector of small dimension (vector of principal components), which is then compared with the reference vectors embedded in the database. The main purpose of the principal components method is to significantly reduce the dimensionality of the feature space so that it best describes the "typical" images that are common to all individuals. Using this method, it is possible to identify various variability in the educational sample of facial images and describe this variability in the basis of several orthogonal vectors, which are called eigenvectors. In almost all algorithms, the obligatory stage that preceding the classification is alignment, which means aligning the image of a person to the frontal position relative to the camera or bringing a set of persons (for example, in a training sample to train a classifier) to a single coordinate system. To implement this stage, it is necessary to localize anthropometric points characteristic of all persons - most often the centers of the pupils or corners of the eyes. Active appearance models are designed to accurately localize these anthropometric points on the face image. Active appearance models are statistical models of images that can be adjusted to a real image by various deformations. The active appearance model contains two types of parameters: parameters related to the shape (shape parameters) and parameters related to the statistical model of image pixels or texture (appearance parameters). Machine learning is the learning of computer algorithms that are automatically improved through experience in analyzing and using data. This is seen as part of artificial intelligence. Machine learning algorithms create a model based on sample data known as "learning data" to make predictions or decisions without being explicitly programmed to do so. Certain machine learning algorithms are called scenarios and differ in the type of training data available to the algorithm, the order and method by which the training data are obtained, and the data used to evaluate the learning algorithm. There are also classic machine learning tasks based on certain machine learning scenarios. For the task of recognizing faces by facial images, the classification task and the learning scenario with the teacher are most often used, because for each set of learning data the exact class to which this set belongs is known, as well as the number of resulting classes is clearly known. Controlled learning, or teacher learning scenario, is a machine learning task that involves learning a function that compares input to output based on examples of input-output pairs. While learning teaching with a teacher, each example is a pair consisting of an input object (usually a vector) and a desired output value (also called a control signal). The learning algorithm with the teacher analyzes the learning data and creates a predictable function that can be used to display new examples. Classification is the problem of determining which of the categories a particular observation belongs to. In machine learning tasks, observations are often called instances, and possible categories that need to be predicted are called classes. The task of classification is often reduced to the analysis and selection of the most optimal parameters of the algorithm to obtain the most desirable parameters for training time, learning accuracy and learning stability. The stability of the algorithm to change the input data is an important parameter, because in the machine learning task it is impossible to teach the model when adding new data and, accordingly, class labels, and it is necessary to retrain the model on a new data set. With low resistance to change, the accuracy of learning the algorithm can vary greatly at constant parameters. | uk |
| dc.format.extent | 75 с. | uk |
| dc.identifier.citation | Бондарев, В. Р. Розпізнавання облич методами машинного навчання : дипломна робота … бакалавра : 153 Мікро- та наносистемна техніка / Бондарев Вʼячеслав Ростиславович. – Київ, 2021. – 75 с. | uk |
| dc.identifier.uri | https://ela.kpi.ua/handle/123456789/59622 | |
| dc.language.iso | uk | uk |
| dc.publisher | КПІ ім. Ігоря Сікорського | uk |
| dc.publisher.place | Київ | uk |
| dc.subject | ідентифікація | uk |
| dc.subject | зображення | uk |
| dc.subject | обробка зображень | uk |
| dc.subject | ознаки особи | uk |
| dc.subject | машинне навчання | uk |
| dc.subject | класифікація | uk |
| dc.title | Розпізнавання облич методами машинного навчання | uk |
| dc.type | Bachelor Thesis | uk |
Файли
Контейнер файлів
1 - 1 з 1
Вантажиться...
- Назва:
- Bondarev_bakalavr.pdf
- Розмір:
- 2.86 MB
- Формат:
- Adobe Portable Document Format
- Опис:
Ліцензійна угода
1 - 1 з 1
Ескіз недоступний
- Назва:
- license.txt
- Розмір:
- 9.1 KB
- Формат:
- Item-specific license agreed upon to submission
- Опис: