Large-Scale Video Retrieval via Deep Local Convolutional Features

Algorithm 1

Key-frame extraction based on the K-means cluster.

	Input: the original video sequence
	Output: a key frame sequence
	(1) Split video data into a set of frame sequences ()
	(2) Calculate the Euclidean distance between adjacent frames according to the color histogram
	(3) Calculate the mean distance
	(4) Assuming the number of key frames is m, affected by the values of parameters and
	(5) for j = 1, …, n − 1 do
	if then
	m + = 1
	end if
	end for
	(6) Select m cluster centers randomly
	Repeat
	(7) Extract deep convolutional features of video frames via VGG16
	(8) Calculate the distance between each frame and the cluster center via the deep convolutional features
	(9) Reclassify the corresponding frames according to the minimum distance criterion
	(10) Recalculate the cluster center of each class
	Until the objects in each cluster no longer change
	(11) The cluster center of each class is available, and the frame closest to the cluster center is selected as a key frame