The Shot Boundary Detection Method Based On DEMD

Nowadays our world is coming into the age of information,along with the development of the computer software andhardware and the popularization of the network; a great deal ofimagesandvideosignalscanbestoredandtransmitted.Wecanreceived a lot of digital images, digital video signals, pictures ofnews, medicinal pictures, pictures of remote sensing and so oneveryday. So how to organize, express, store, manage, query,search the large number of data is a momentous challenge totheconventionaldatabase.Aconventionaldatabaseistakingthecharacter as the object, can not satisfy the request of imagedatabase. If we can not give the image and video data aautomatic and effective description, a lot of information will besubmerged in the database, even can not be searched whenthey are needed. So how to make the computer visiontechnology, digital image processing and conventional databasetechnology together into a new image video database which isbased on image and video data isa practical investigation task.Content based retrieval, namely CBR is a technology ofsearch which is based on kinds of character, and the charactercan describe the object in medium. CBR can find the image orthe video which has the appoint characters from the database.CBR is different from the traditional search method which isbased on the keyword, mixed with many technologies such asthe image understanding technology, pattern recognitiontechnology.In the content-based video retrieval system, namely CBVR,there are the following steps. Firstly, through the analysis of thestructure of the video stream, we can segment the videosequence into some shot, then take a key frame to describe themain content in the shot. The first step is the base andimportance of a high efficiency CBVR. Secondly, we extract thekey frame and the movement features in each shot, and thenstore them into the video database as a kind of retrievalinformation. Thirdly, according to the query of user, use somefeatures to search in the database, then feedback the result tothe user with a similarity. When the user is not satisfaction withthe retrieval result, we can optimize the result with the user'sopinion.Conventional, we can use a hiberarchy model to describe avideo. The hiberarchy model is made of scene, shot and frame.Shot is the basic unit in the video edit, video index, and videoquery. A shot is a series of correlativity frames which are takingfrom a vidicon'continuum shoots. We must divided the videosequences into some shots, the following works can be welldone, such as the extracting of key frames, the compressing ofvideo, the identifying of video sequences. In the CBVR, thekernel technology is the detection of the shot boundary. Toidentify the first frame and the last frame in a shot, to identify theshot transition type (cut or gradual), even to identify the type ofthe gradual (dissolve, fade-in, fade-out or wipe) is the content inthe detection of the shot boundary. Now the detection of the cuttransition has got a well production, but the detection of gradualtransition still is a question for discussion, because of itscomplicacy.The approach for shot boundary detection can be classedinto three types, namely pixel-based method, block-basedmethod, histogram-based method. The pixel-based method isthe easiest method to measure the discontinuousness in thevision, but this method is sensitivity to the global motion andlocal motion, and the amount of calculation is very large. Thehistogram-based method takes the histogram of each frame asfeature. This method thinks the sequential frames in one shothave the same global vision feature, and this feature isrepresented as the differences in histograms between theseframesissmallerthanthe differencesinhistogramsbetween thetwo frames at the shot boundary. Because of thehistogram-based method take the attention of the globaldistributing, this method is not sensitivity to the local movement.But when there is a global movement in the shot, the histogramwillhaveagreatchange.Thehistogram-basedmethodhasabigdisadvantage, when two video images have different structure,the histograms maybe very similar, this will result in the missdetection. The block-based method is proposed to resolve thedisadvantage of histogram-based method, but this method isverysensitivityto the local movement in theshot.The above-mentioned methods all have a disadvantage,they all very sensitivity to the illumination break. Through theresearch of EDM, we find use directional EMD, namely DEMDcan resolve the problem of illumination break in the shot.Illumination break only leads the lowest frequency IMF imagehave great changes in a video shot, this is because the lowestfrequencyIMFcomponentrepresentstheaverageortrendoftheoriginal signal. To the image signal, the lowest frequency IMFcomponent represents the illumination distribution and energy inthe image. So in our algorithm, we use DEMD decompositionmethod to resolve the wrong detection because of theillumination break in the video shot. Firstly use the DEMDmethod to deal with the video frames, after this processing wecan get the new video frames which are eliminated theillumination break to replace the old video frames. Then,compute the histograms on the new video frames. This methodresolves the sensitive of illumination break in traditional method.The object movements and the shoot movements exist ineveryshot sequence. No matterwhich similarityfeatures we use,the movements in the shot will lead the sequence frames have abigdifference,andthenleadthewrongdetectioninthedetectionof shot boundary. So motion compensation is very important inthe detection of shot boundary, we use a simple and efficientmethodtodothemotioncompensation.Weuseanaccumulativeframe C to accumulate the differences between the blocks in Cand the corresponding blocks in the successive frames.According to the C, we can do the motion compensation, andthen make the detection of shot boundary become not sensitiveto the object movement and shoot movement.In our algorithm, we use the DEMD method and anaccumulative frame C to resolve the disadvantage in thetraditional method. The algorithm can resolve the influence ofillumination break and the sensitive to the object an shootmovement, then reduce the wrong detection and the missdetection in the detection of shot boundary. Our algorithm canenhancethedetectionof shotboundary'sprecision and keep therecall's effect.