Research On Algorithms For Video Feature Mining Based On Fuzzy Concept Lattice

With the rapid development of computer and multimedia technologies, multimediadata, especially video data, increases tremendously. The video is one of the importantsources to get information, which contains much more information and is more intuitivein expression compared with the static image. How to extract the features of the video,such as texture, shape and motion feature etc., from massive video data rapidly andeffectively has become a hotspot and difficulty in video processing. Extended from theimage extraction algorithms, the existing algorithms to extract the features of the videocan not well meet the real-time requirement in video processing. To solve the aboveproblems, the main research of this paper includes the following aspects.Firstly, a method based on fuzzy concept lattice to mine the texture feature of thevideo is proposed. The fuzzy concept lattice and gray level co-occurrence matrix arecombined. The gray level co-occurrence matrix is used to describe the texture feature ofthe video from different perspectives, containing energy, entropy and contrast etc, whichare the attributes of the fuzzy formal context. The texture association rules are generatedby the fuzzy concept lattice, and the video texture feature is mined rapidly based on theassociation rules. Experimental results show that this method can mine the texturefeature of the video rapidly and saves a lot of computation compared with the existingmethod and meets the real-time requirement in massive video processing.Secondly, a method to mine the motion feature of the video based on fuzzyconcept lattice is proposed. By this method, the fuzzy concept lattice and the MPEG-7motion activity descriptor are combined. The MPEG-7motion activity descriptors areused to describe the motion features of video shots and video frames from differentperspectives, which are the attributes of the fuzzy formal context. The motionassociation rules are generated by the fuzzy concept lattice and the shots of interest, theframes of interest and the motion features of the interested frames are mined rapidlybased on the motion association rules. Experimental results show that the shot ofinterest and the frame of interest mined by this method are consistent with the humanperception, and that the motion feature of the interested frame mined by this method canreduce the influence of the background noise, which prove the rapidness andeffectiveness of the proposed method. Finally, the research contents of this paper are summarized and the next researchdirection is propounded.