3.1.1 Experimental Result
To evaluate our slide-video segmentation method, we have
compiled a lecture video test data set, which consist of 20
randomly selected videos from different speakers with various
layouts and font styles. Each unique slide with complete
contents is regarded as a correct match. The overall number
of detected segments is 860, of which 816 are correct. The
achieved segmentation recall and precision are 98 and
95 percent for the test data set, respectively. Since this experiment
is designed to evaluate the slide segmentation algorithm,
videos with various genres embedded within the
slides are not considered. The detailed evaluation result as
well as the url for each test video can be found at [21].
To evaluate the slide-image classifier, 2,597 slide frames
and 5,224 non-slide frames have been collected in total for