VISON: VIdeo Summarization for ONline applications

Video Key Frame Extraction through Canonical Correlation Analysis and Graph Modularity

“…Specifically, most video codecs (i.e., MPEG-1/2/4) are based on group of pictures(GOPs) as basic units [6]. The content of a GOP is represented using I-frames.…”

“…The content of a GOP is represented using I-frames. The compression of the I-frames of a MPEG video is carried out by dividing the original image into 8 x 8 pixel blocks and transforming the pixels values of each block into 64 DCT coefficients [6]. Finally, on extracting the DC term of all the pixel blocks, a reduced version of the original image (known as DC image) is produced whose size is only 1/64 of original image.…”

“…Long videos are avoided due to limitation of annotation by a subject. The test set has also an intersection with the work in [5,6]. All the experiments are performed on a machine with Intel(R) core(TM) i5-2400 processor and 8 GB of DDR2-memory.…”

“…Moreover, many of those research works have focused on the uncompressed domain which makes the system unsuitable for online usage. Although the work presented in [6] is in compressed domain it utilizes the notion of similarity between successive frames. However, choice of similarity measures greatly influences the effective content representation of the key frame set.…”

“…However, choice of similarity measures greatly influences the effective content representation of the key frame set. Most of the approaches use a single visual feature, i.e., color to represent a video frame [5][6]. However, color alone cannot in general capture all the pictorial details needed to estimate the changes in the visual content of frames.…”