Personalized Sports Video Customization Using Content and Context Analysis

Table 3

Sequence matching algorithm for video-text alignment.

Output	Alignment sequence 𝐚∶𝐚1∼𝐚max(𝑚,𝑛)
Input	Video tag sequence 𝐯∶𝐯1∼𝐯𝑚 and text tag sequence 𝐭∶𝐭1∼𝐭𝑛
Step 1	Forward computation
Step 1.1	Allocate a score matrix 𝙼 and a direction matrix 𝙱.
	Both matrixes are with 𝑚 rows and 𝑛 columns.
Step 1.2	Fill all entries in 𝙼 and 𝙱 from the up left to bottom right corner.
	(i) For each 𝙼𝑖,𝑗, its value is calculated by (3), where 𝑠𝑖,𝑗 and Pg are computed by (4) and (5).
	(ii) For each 𝙱𝑖,𝑗, its value is the optimal matching direction indicated by 𝙼𝑖,𝑗 in (3).
Step 2	Backward tracing
Step 2.1	Allocate an alignment sequence 𝑎 with max(𝑚,𝑛) elements.
Step 2.2	Fill the entries of 𝐚 in a reverse direction by tracing back along 𝙱.
	(i) For each 𝐚𝑘, its value is equal to 𝙱𝑖,𝑗, where (𝑖,𝑗) corresponds to the coordinate of the 𝑘th backtracking step in 𝙱.
Step 3	Result explanation
Step 3.1	Three possible values (directions) may appear in 𝐚𝑘, which are
	(i) diagonal, corresponding to the “𝐯𝑖−𝐭𝑗” alignment,
	(ii) downward, corresponding to the “empty-𝐭𝑗” alignment,
	(iii) rightward, corresponding to the “𝐯𝑖-empty” alignment.