12.07.2015 Views

an innovative algorithm for key frame extraction in video ...

an innovative algorithm for key frame extraction in video ...

an innovative algorithm for key frame extraction in video ...

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

energy feature, the magnitudes of the motion vectors of the B-<strong>frame</strong>s are firstfiltered with two nonl<strong>in</strong>ear filters. For each motion vector <strong>in</strong> the <strong>frame</strong> feature, aspatial filter is applied with<strong>in</strong> a given spatial w<strong>in</strong>dow, <strong>an</strong>d a temporal filter isapplied on values belong<strong>in</strong>g to <strong>frame</strong>s with<strong>in</strong> a given temporal w<strong>in</strong>dow. For eachB <strong>frame</strong>, the PME is then computed on the magnitudes of the motion vectors <strong>an</strong>dthe dom<strong>in</strong><strong>an</strong>t motion direction. This preprocess<strong>in</strong>g requires the sett<strong>in</strong>g of severalparameters.A simple procedure then automatically computes the tri<strong>an</strong>gles on the PMEvalues <strong>an</strong>d the correspond<strong>in</strong>g <strong>key</strong> <strong>frame</strong>s. The <strong>algorithm</strong> requires the sett<strong>in</strong>g oftwo parameters the most import<strong>an</strong>t of which is the m<strong>in</strong>imum size of a tri<strong>an</strong>gles<strong>in</strong>ce it <strong>in</strong>fluences the length of the <strong>in</strong>terval between two consecutive <strong>key</strong> <strong>frame</strong>sThe MP <strong>algorithm</strong> was chosen because it c<strong>an</strong> represent the extreme case of our<strong>algorithm</strong>, when no evident high curvature po<strong>in</strong>ts c<strong>an</strong> be found <strong>in</strong> a shot <strong>an</strong>d thecenter <strong>frame</strong> of the sequence is chosen as the <strong>key</strong> <strong>frame</strong> <strong>in</strong>stead.Where available the parameters set <strong>for</strong> the <strong>algorithm</strong>s were always thosereported <strong>in</strong> the orig<strong>in</strong>al papers. The ATS, FR <strong>an</strong>d SRDI <strong>algorithm</strong>s require the<strong>in</strong>put parameter of the number of <strong>key</strong> <strong>frame</strong>s that must be provided. Def<strong>in</strong><strong>in</strong>g ageneral rule <strong>for</strong> sett<strong>in</strong>g this number is a crucial matter; mthe results may varywidely, depend<strong>in</strong>g on the rule selected. We have set the <strong>in</strong>put parameter <strong>for</strong> these<strong>algorithm</strong>s as the same number of <strong>key</strong> <strong>frame</strong>s found by our <strong>algorithm</strong>. We c<strong>an</strong>then compare the <strong>algorithm</strong>s regardless of the number of <strong>key</strong> <strong>frame</strong>s: <strong>an</strong>ydifference <strong>in</strong> results depend only on the selection strategy adopted. S<strong>in</strong>ce the PME<strong>algorithm</strong>, <strong>in</strong>stead, extracts the <strong>key</strong> <strong>frame</strong>s <strong>in</strong> a totally automatic way, as does our<strong>algorithm</strong>, the results depend on both the number of <strong>key</strong> <strong>frame</strong>s extracted <strong>an</strong>d theselection strategy applied.5.2 Video Data SetSix <strong>video</strong>s of various genre were used to test the per<strong>for</strong>m<strong>an</strong>ce of the <strong>key</strong> <strong>frame</strong><strong>extraction</strong> <strong>algorithm</strong>s. Table 1 summarizes the characteristics of the six <strong>video</strong> testsequences. The “eeopen” <strong>video</strong> is a MPEG1 <strong>in</strong>tro sequence of a TV series withshort shots <strong>an</strong>d several tr<strong>an</strong>sition effects. The “news” <strong>an</strong>d “nw<strong>an</strong>w1” are twoMPEG1 news sequences; the shots are moderately long, not too dynamic, <strong>an</strong>dmixed with commercial sequences of very fast-paced shots. The “nw<strong>an</strong>w1” <strong>video</strong>is similar to the “news” <strong>video</strong>, but has longer shots. The “football” <strong>an</strong>d“basketball2” <strong>video</strong> are two MPEG1 sport sequences: “football” exhibits ratherlong shots, while “basketball2” is a s<strong>in</strong>gle long shot, <strong>an</strong>d both have p<strong>an</strong>n<strong>in</strong>g <strong>an</strong>dcamera motion effects. F<strong>in</strong>ally, “bugsbunny” is a MPEG1 short cartoon sequencewith m<strong>an</strong>y shots <strong>an</strong>d a number of tr<strong>an</strong>sition effects.Table 1. The six <strong>video</strong>s used to test the <strong>key</strong> <strong>frame</strong> <strong>extraction</strong> <strong>algorithm</strong>s. TNF denotes the totalnumber of <strong>frame</strong>s, <strong>an</strong>d NS, denotes the number of shots found; both refer to type A <strong>an</strong>d type Bshots.Video Name GenreLength Resolution(mm:ss) (W×H)TNF NSeeopen TV series <strong>in</strong>tro 00:42 352×240 1 289 24nw<strong>an</strong>w1 News with commercials 03:39 176×112 6 556 39news News with commercials 02:39 176×112 4 757 12football Sport 03:43 176×112 6 697 28bugsbunny Short cartoon 07:30 352×240 13 492 89basketball2 S<strong>in</strong>gle shot sport sequence 00:30 320×220 893 115

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!