Content-Based Video Browsing: semantic similarity and personalization Jamel SLIMI, Anis Ben AMMAR, Adel M. ALIMI REGIM-Lab.: REsearch Groups in Intelligent Machines, University of Sfax, ENIS, BP 1173, Sfax, 3038, Tunisia. Abstract-In this paper, we present an intelligent video browsing system covering all tasks in video data visualization process. Visualization process is composed by categorization step followed by a representation of video collection step. The specificity of our work resides in the integration of personalization module allowing an appropriate interface to the user preferences. Our tool is based on multimodal video indexing (video text extraction, audio features and visual features). Video Indexing allows the construction of video data descriptor vectors. Based on these vectors, we calculate semantic similarity distance between documents composing video collection. This task permits a semantic classification of video corpus. Obtained classes will be projected in the visualization space. Video data visualization graph is in the form of a network. This network is composed by nodes (keyframes extracted from video shot) and color edges representing the similarity distance between data collection. Visualization interface components comportment is inspired from biological neuron comportment. By clicking on keyfarme representing document; all the documents which are strongly connected to this one will be posted in the visualization space. An important step in our tool is dedicated to integrating personalization module in the video data visualization system. Personalization is based on user preferences collection. These preferences are collected via user interaction with the system. User profile is based on static indicators, dynamic indicators and navigation history. Compared to existing video browsing; our system includes a personalization module allowing appropriate interface to the user preferences. Network form of visualization representation permits easier navigation in large video corpus. Key words: data visualization, video semantic similarity, personalization, video indexing, content-based video browsing. I. INTRODUCTION Multimedia data provide large amounts of information for users. They cover most of the daily events; and they assure the awfully various requests of the people (sport, cultural, political...) [1]. This wealth of information is based on the perceptual human appearance. In fact, an image can expressed as 1000 words. Interest in the treatment of this type of data is caused by the large amount of information generated daily [2]. This increase in production, especially for video data, is due to the technological progress in video production and the availability of digital cameras [3]. Therefore, the size of the video data collections has become very large, either in personal database or in the database of dedicated Web sites [4]. It generates the birth of several evaluation companies as TRECVID which is interesting to several research areas taking video data as a data source such as indexing, classification, concepts extraction, etc. These were well treated by several researchers. Consequently, the subject of exploration and navigation in multimedia databases is not yet considered in the right way. In fact, a simple exploration and fast access to documents in the large video corpus has become an urgent need. Video data visualization systems need to overcome several problems. The first problem resides in the difficulty of the video semantic concepts extraction. Semantic description of the video data offers improved classification International Journal of Computer Science and Information Security (IJCSIS), Vol. 14, No. 7, July 2016 706 https://sites.google.com/site/ijcsis/ ISSN 1947-5500