Proceedings of National Conference on Emerging trends in VLSI, Embedded and Nano Technologies, January 27 - 28, 2011.

VIDEO DATA MINING FRAMEWORK FOR INFORMATION RETRIEVAL SYSTEMS

D.Saravanan Research Scholar Sathyabama University Chennai-119. 1. Abstract - Data mining technique can be applied in various documents. In this paper concentrationon the application of video data, called video data mining, because acquisition and storage of video data is an easy task but retrieval of information from video data is a challenging task. So video data mining plays an important role in efficient video data management for information retrieval. This paper describes a proposed framework for video data mining to extract the information from video data. This includes developing the technique for shot detection then key frame analysis is considered to compare the frames of each shot to each others to define the relationship between shots. After all hierarchical clustering technique is adopted to make a group of similar shots to detect the particular event on some requirement as per user demand. Key Words- Video data mining, key frame analysis, clustering technique. Dr.S.Srininvasan Head, School of Computer Science Tamil Nadu Open University Chennai. mining directly [1] [2] [3]. In video databases, knowledge discovery deals with nonstructural information, for this reason we need tools for discovering relationships between objects or segments within video components. In general video must be first preprocessed to improve their quality. Subsequently, these video files undergo various transformations and features extraction to generate the important features from the video databases. With the generated features, mining can be carried out using data mining techniques to discover significant patterns. These resulting patterns are then evaluated and interpreted in order to obtain the final applications knowledge. The problem of video data mining combines the area of content-based retrieval, image understanding, data mining, video representation and databases [4] [5] [6]. Many applications maintain temporal and spatial features in their databases; these features cannot be treated as any other attributes and need spatial attention. To be more particular instead of simply asking ourselves what knowledge, it plays an important role and these can be trivially handled by spatial and temporal role and these can be trivially handled by spatial and temporal data mining techniques [7]. Finally data mining of video databases is essentially a task of learning from video data can, in principle, being applied for data mining purposes. In general data mining algorithms aim at minimizing I/O operations of disk resident data, whereas conventional algorithms are more concerned about time and space.

1. INTRODUCTION A wide range of possible applications that require data mining of video databases includes: news broadcasting, military, video, education and training, cultural heritage, advertising, web searching, crime prevention, geographical information system (GIS) etc. these applications has vast collection of images in the corresponding video databases and can be mined to discover new and valuable knowledge. Data mining of video databases aims to automate such a knowledge discovery process. To help user finds and retrieves relevant video effectively and to facilitate new and better ways of entertainment, advanced technologies must be developed for searching and mining the vast amount of videos now available on the web. Although valuable information may be hiding behind the data, the overwhelming data volume makes it difficult for human beings to extract them without powerful tools. Video mining system that can automatically extract semantically meaningful information (knowledge) from video databases. While numerous papers have appeared on data mining, few deals with video data

Fig 1. Video Data Mining Model

10

Proceedings of National Conference on Emerging trends in VLSI, Embedded and Nano Technologies, January 27 - 28, 2011.
Simply stated, the objective of video data mining is the organizing of video data for knowledge exploring or mining. Where the knowledge can be explained as special patterns (e.g., events, clusters, classification, etc.), which may be unknown before the processing. Many successful data mining techniques have been developed through academic research and industry hence, an intuitive solution for video data mining is to use these strategies on video data. 2. REQUIREMENTS OF A VIDEO MINING SYSTEM Following requirements for a video mining system: 1. It should be unsupervised. 2 It should not have any assumptions about the data 3. It should uncover interesting events 4. It should not have any assumptions about the data. 5. It should uncover interesting events Note that requirements 2 and 3 are somewhat contradictory, since the notion of interesting is subjective, and highly dependent on knowledge of the content. The range between purely unsupervised and purely supervised techniques can be thought of as a continuum that goes from the general to the particular. Our aim is to find out how few assumptions we can make about the content without detecting events that are too general to he meaningful. This would help us understand the content specific heuristics reported in most previous work, and help set up a framework for systematic use of domain knowledge. Note that sports video is in our view a good genre to start with because it has some constraints hut is spontaneous as well. It thus provides a tractable test-bed for pattern discovery techniques. 2.1 Image Mining Algorithm Steps: The four major mining steps are as follows: 1. Feature Extraction: Segment image into regions identifiable by region descriptors. Ideally one descriptors represents one object. This step is also called segmentation. 2. Object identification and record creation: Compare objects in one image to objects in every other image. Label each object with an id. We call this step the preprocessing algorithm. 3. Create auxiliary images: Generate images with identified objects to interpret the association rules obtained from the following step 4. Apply data mining algorithm to produce object association rules. 3. STEPS OF PROPOSED FRAME WORKS: Figure 1 depicts the frame work of video data mining consists of several steps, primary steps is segmentation of a video, because we can not apply the data mining directly to retrieve the information from video or detect the pattern from video data. So division of video into small segments is done basically to support relational database. Then further key frame analysis and clustering technique are involved to analysis the video data as well as retrieve the information.

Fig 2. Video Retrieval Systems 3.1 SHOT DETECTION TECHNIQUE The primary step of video data mining is to segment the video into small shots to represent into relational data format. Video segmentation, or shot change detection, involves identifying the frame(s) where a transition takes place from one shot to another. In cases where this change occurs between two frames, it is called a cut or a break. A shot is defined as a sequence of frames taken by a single camera with no major changes in the visual content. Much technique has been developed to detect the shots. The most common approach to obtain boundaries of a shot is based on color information of the frame, [8] [9] [10]. In the recent studies, no matter which color space is of interest, main advantages of using color is its ease of implementation, descriptive characteristic both in spatial and temporal space and real-time applicability due to simplicity to obtain a feature vector, such as histogram [11]. In this way read the total no. of frames in the given video, after all divide the video into the shots on the basis of color observation of the frames.

11

Proceedings of National Conference on Emerging trends in VLSI, Embedded and Nano Technologies, January 27 - 28, 2011.
particular instance of the collection of objects, and reflects how alike the objects are in some interesting way. Clustering can be used either a stand-alone tool to get insight into the data distribution or as a preprocessing step for other. Clustering is often an important initial step of several in the web data mining process. 4 .SOME RESEARCH ORGANIZATIONS OF VIDEO MINING Based on our investigation, to date, the following organizations have done deeper work on video mining. (1) DIMACS Workshop. DIMACSI play a national leadership role in the development, application and dissemination of the inteiTelated fields of discrete mathematics and theoretical computer science. They held an international conference on video mining on November 4-6, 2002 at Rutgers University. (2) MIERL-. The researchers consider the video mining problem in the light of existing data mining techniques their approaches to video mining are to think of it as "blind" or content-adaptive processing that relies as little as possible on a priori knowledge. The processes of video mining are first to extract the features of video content, and then use adaptively constimicted statistical models of usual events to discover unusual audio-visual pattems. (3) DVMMlab4 of Columbia University. They give attention to mining periodic patterns from semantic video events from 2001. They think video patterns are multilevel, from features to events, which can be discovered by using hierarchical hidden Markov models, 5. CONCLUSION AND FUTURE WORK In this work a frame is proposed for video data mining where it can be applied for many video applications to detect or retrieve the information. So this is the common process for video data miming. Future research work is to take a particular application and apply this frame work for video data mining. The main goal of this work is to develop the intelligence technique that can be used to support the video data mining process to retrieve information successfully and maintain the accuracy of the system. REFERENCES [1] J. Fan, X. Zhu and X. Lin, Mining of video database. Book chapter in Multimedia data mining, Kluwer, 2002.

Fig 3. Video Mining System Arch. 3.2 KEY FRAMES ANALYSIS After detection of the shots from given medical video, key frame analysis is required to extract the features and define the relationships between the frames in shot. Choosing key frames of shots allows us to capture most of the content variations, due at least to camera motion, while at the same time excluding other key frames which may be redundant. The ideal method of selecting key frames would be to compare each frame to every other frame in the shot and select the frame with the least difference from other frames in terms of a given similarity measure 3.3CLUSTERING TECHNIQUES INFORMATION RETRIEVAL AND

Clustering similar shots into one unit will eliminate redundancy and produce a more concise video content summary. In unsupervised classification, the problem is to group a given collection of unlabeled video files into meaningful clusters according to the video content without a priori knowledge. Clustering algorithms can be categorized into partitioning methods, hierarchical methods, density-based methods, grid based methods, and model-based methods[11]. For this work hierarchal clustering clustering is adopted to represent the cluster of the similar shots. An excellent survey of clustering techniques can be found. When clusters are generated to define the similar shots then events mining strategies is considered to evaluate the shots as per user demand. This technique is based on rules or visualization of the shots. After processing of the visualization on such rules particular information in form of shot is detected to achieve the information retrieval from video data. A clustering is a process of partitioning the data in meaningful groups called clusters, and it is used in help understand the natural grouping or structure of a data set. A cluster in a collection of objects that are similar to one another. Similarity here is defined based on the

12

Proceedings of National Conference on Emerging trends in VLSI, Embedded and Nano Technologies, January 27 - 28, 2011.
[2] J.-Y. Pan, C. Faloutsos, .Video Cube: A new tool for video mining and classification. ICADL Dec., 2002, Singapore. [3] S.C. Chen, M.L. Shyu, C. Zhang, J. Strickrott, .Multimedia data mining for traffic video sequence., MDM/KDD workshop 2001, San Francisco, USA. [4] Stephane , Marchand- Maillet , content- based video retrieval: an overview. [5] Matthias Kreuseler & Heidrun Shumann. A flexible approach for visual data mining, IEEE 2001. [6] U.M.Fayyad , G. Piatetsky- Shapiro, P. Smyth , and R.Uthurusamy, advanced in knowledge discovery and data mining. AAAI/MIT press, 1996. [7] Raymond T. Ng and Jiawei Han . member,IEEE computer society. CLARANS: A method for clustering objects for spatial data mining . IEEE transactions on knowledge and data engineering, vol.14, No.5, septoct.2002. [8] W. Zhao, J. Wang, D. Bhat, K. Sakiewicz, N. Nandhakumar and W. Chang, Improving color based video shot detection, IEEE Int. Conf. On Multimedia Computing and Systems, Vol. 2, 1999, pp. 752-756. [9] R. Zabih, J. Miller, and K. Mai, A feature-based algorithm for detecting and classifying scene breaks, Proc. ACM Multimedia 95, San Francisco, CA, November, 1993 , pp. 189-200. [10] E. Sahouria, A. Zakhor, Content Analysis of Video Using Principal Components, IEEE transactions on circuits and systems for video technology, Vol. 9, No. 8, December 1999, pp. 12901298. [11] M.J. Swain, D.H. Ballard, Color indexing, Int. Journal of Computer Vision, Vol. 7. 1991, pp. 11-32.

13