Social networking has now become the primary means of communication, support and development of social contacts, searching, storing, editing, and classifying information, creativity and perform many other tasks. Despite the diversity of views and preferences of people in the network there is a repetition of the published material. This is due primarily to the taxonomic and folksonomic classification data.

Lack of taxonomic approach is that the object can be bound in such only to one node, so it's impossible with such a structure to describe all the necessary qualities of the object. In this regard, it is obvious copying of the same object in another node with a description of other qualities.

Taksonomic approach does not have this drawback - you can bind an object to any sites. However, in the latter approach, the absence of any structure, ie lack the basic relationship (genus-species) between nodes. Thus, it is impossible to identify objects of a more general or more specific. It is also a significant drawback for video hosting services with huge amounts of information, which reduces the efficiency of search and leads to the creation of copies of files. [5][4]

It should be noted that the number of copies of the video posted on a video service or a social network depends on the popularity of video and fashion trends. Of personal research has revealed that a "popular" video posted on a video service from 5 to 20 times, and only 1-2 of them are different video quality, video, and large amounts of video network services is a significant amount of disk space, measured in terabytes. Also increases the amount of storage using RAID technology on 10 servers, and CDN-network content delivery. [5]

The aim is to reduce storage costs and increasing the speed of the video sharing server, by identifying similar video material. To achieve this goal in the research process should

1. Consider and explore the existing methods to search for identical video;
2. Perform analysis of video compression algorithms;
3. Develop an algorithm to capture and convert the video into a sequence of frames;
4. Develop an algorithm for recognizing and classifying the obtained sequences.
5. Appreciate the complexity of the implementation of the developed system, and identify areas for its effective application

Relevance work is determined by popularity of social networking and video hosting. The developed system will is much reduce the amount of storage. And delete duplicate video is one of the most pressing problems of optimization video hosting.

Probable scientific novelty

To search for identical videos on video hosting is used hash or checksums. These processes take a long time, so once the hash are calculated - the amount savedf in database for future use. This method can not analyze video sequence data and is unable to determine the equivalent in content but different in size, codec, compression, file permissions.

The developed system allows under data obtained from video sequences to compare the videos using system pattern recognition, and making assumptions about the degree of similarity of the video. The system has a completely new algorithm for identifying the video, those with not have eqval in the CIS and the international community.

In the master's work is planned to develop an automated system for pattern recognition. Its main tasks will be:

• Creating a set of initial data
• Presentation of the original data set obtained as a result of measurement for the object to be recognition
• Classification and identification of an object by using the optimal decision procedure
• Find and delete video files identified as equivalent or very similar

The result will be a web-based application, which will be implemented using several technologies:

1. PHP - are using this technology for building server pages with dynamically forming information stored in the database.
2. MYSQL - database that will store the results of the work and the results classification objects.
3. JavaScript - will be used to enhance your user interface system
4. C + + - develop a library that will be executed on a Web server using library Appache FastCGI. Using the compiled code will much faster system performance.

On the results of the search materials on portal Masters DonNTU were found working on similar tasks, but, but among them as the chosen methods of solution, and the scope of this work are very different. They have developed: Isayenko A., Driga K., Sova A..

Though in Ukraine and at the moment have not been developed production systems for pattern recognition. In the Russian Federation of the given market IT solutions. Many experts and different companies develop systems for pattern recognition for solving various problems, special attention is deserved: A. Vakhitov with the development of video monitoring systems, the company, "Mullen" with the development of "Video monitoring in transport"

Abroad development of pattern recognition are developing large companies as a: Philips, Sony, Samsung, Lexus, Toyota, Siemens. In developed systems differ greatly applications, the main goal of systems reducing errors associated with the human factor and solution of problems previously solved only by man.

Formation the task set of source data

This task is the initial stage of the system. At this stage the capture of the movie in a series of specified time intervals at the end scenes (Fig. 1).

Figure 1 - Capture video frames

Then the captured image is reduced to a single solution, such as 700x400 (Fig. 2a). This is necessary because different videos can be cut, reduced or enlarged by third parties or their resolution may not match the worker. Next frame is translated into shades of gray (Fig. 2b) should be converted to gray is as a a space-saving storage of measurement results (single-pixel image can be encoded a byte of information from the gray value 0 .. 255), so and to exclude the possibility of using different color spaces encode the video. then frame is normalized (Fig. 2c). and smeared Gaussian blur algorithm (Fig. 2d). These actions correct as a compression artifacts and the alignment of non-uniformly distributed levels of the image.

Figure 2 - A Cutting-700x400, B-translation in grayscale, C - normalization image, D- Gaussian blur

The task of providing input

The task of providing input, meant to receive recognition of measurement results to be an object. Each measured value is a "characteristic" of an image or object. In this system, a frame consists of 64 images. Each image is 1 / 64 of the evenly divided image. The process of separating the image shown in (Fig. 3).[1][2][3]

Figure 3 - The process of obtaining a sample
Animation: size - 100 KB; size - 216x173, the number of frames - 50;
cycles repetition - the infinite

this case, the sensor can be used successfully measuring the retina, like that shown in Figure 4. If the retina consists of a matrix (m, n) elements, the measurement can be represented as a matrix image. [1]

Figure 4 Presentation of results of measurements

where each element of the Xmn, taking, for example, to [0,255] (1 byte). As mentioned above conversion to grayscale pixel image allowed to encode a byte. [6][7]

Classification and identification of an object by using the optimal decision procedure

Once all the data collected in all classes and are to be collected on pattern recognition, represented by points. Perform the matching algorithm images (Fig. 5). If the comparison of the coefficient of compliance was more than 0.7 can be argued that these videos are the same. [10][12]

Figure 5 Matching algorithm frames

The high degree of duplication of videos, which are located on servers that leads to their excessive workload. Thus, the task of detection and removal of similar videos is important. It is suggested that a comparison of video to make the most logical similarity by comparing frames taken from videos of these specially converted, such as: transfer to the gradations of gray, normalization and filtering of images, in order to increase recognition accuracy. Groups of methods that can solve this problem. It is proposed to use a method that uses the theory of pattern recognition, based on a study of the properties divided by the image. The proposed solution algorithm will significantly reduce the amount of disk space, thereby allowing companies to reduce costs video hosting service providers that will enhance their effectiveness.

1. Распознавание образов - применение на практике.
2. Сайт о распознавании образов
3. Распознавание образов и анализ сцен
4. Архитектура YouTube
5. Википедия. Краткая информация об многих технологиях
6. Pattern Recognition. Finding and Recognizing Patterns in Data
7. Системы распознавания образов
8. Image-Based Face Recognition Algorithms
9. Лапонина О.Р. Криптографические основы безопасности. Лаборатория знаний, Интернет-университет информационных технологий. М.: Бином, 2009. — 536 c.
10. Рутковская Д., Пилиньский М., Рутковский Л. Нейронные сети, генетические алгоритмы и нечеткие системы. М.: Горячая линия -Телеком, 2006. - 452 с
11. Колерс П., Мюррей Д. Распознавание образов. М.: Мир, 1970. - 288 с.
12. Эдвард А. Патрик Основы теории распознавания образов. М. : "Советское радио", 1980.- 864 с.