VIDEO DATA MINING
Department of Computer Science
Jaypee Institute of Information Technology, NOIDA, U.P.
In recent years, advanced digital capturing technology has made digital data grow rapidly. Knowledge discovery from massive amounts of multimedia data, so-called multimedia mining, has been the focus of attention over the past few years. Typically, a video can be viewed as a series of images, which contains a lot of various concepts. Even though the annotation for the individual image frame is in effect, any concept in the image frames still cannot ...view middle of the document...
The database is a rich collection of videos where each video has been divided into key frames. The user enters his query clip broken into individual frames.
An indexing technique is used to store the key frames of all the videos in the database. The frames are then searched to provide the most optimum match of the query clip found in the database. The search is re-ranked when there is more than one video that best matches the query clip in content.
The proposed method contains of two major functions 1) Indexing of videos 2) Searching of videos. The content of the research paper are the review of previous work in this field in section 2. In section 3, we explain our proposed method for video data mining. Finally, conclusions and future work are elaborated in section 4.
2. PREVIOUS WORK
The previous work for video data mining can be categorized into the following based on the extracted visual features, such as color, shape and texture.
1. Key – Frame – Based Retrieval:
Videos are searched by sequential comparisons between the key-frames of the query video and that of the videos in the multimedia repository. Clearly, the computation cost is so high that the users cannot put up with the long response time. Besides computation cost, what seem to be lacking in this paradigm are the considerations for the temporal, sequence and duration of shots in a video.
2. Graph – based – retrieval:
On the basis of the temporal continuity of shots, researchers adopted similarity measures Optimal Mapping (OM) and Optimal Mapping with Replication (OMR) to determine the similarity between two shots. Furthermore, Maximum Matching (MM) was utilized to filter the irrelevant shots and OM was utilized to rank the similarity of clips according to visual and granularity factors. The main disadvantage is that the computation complexity increases rapidly as the amount of videos raises.
3. Pattern mining and image/video analysis:
Recent data mining research has developed many efficient methods for mining frequent patterns, sequential pattern and structural patterns. Such pattern analysis could benefit image and video analysis since similar images and videos may contain some recurrent scenes, components, objects, and video clips. Such recurrent components may disclose important scenes, objects, sequences and structures, and facilitate image analysis. Pattern mining has been recently applied successfully to some image and video analysis tasks.
4. Image/video indexing and similarity search by data mining:
Images and videos can be viewed as graphs with objects/scenes as nodes and their links as edges. The recent developed of graph indexing and its subsequent similarity search techniques based on mining discriminative frequent sub-graphs and using them as primitives makes it efficient to do indexing and similarity search in massive graphs. Also, bag of visual word-based method are also interesting for content image/video retrieval. These...