This article presents an approach for retrieving scenes of interest in long-duration videos through image-to-text encoding. Unlike conventional approaches that often involve the use of neural networks, this method proposes a technique that avoids the use of these complex structures in order to reduce computational resource consumption. Through experiments, the feasibility and effectiveness of this technique are demonstrated, concluding that it is feasible to employ it for multimedia information retrieval, offering an efficient and economical alternative for this task.

Information retrieval, scene identification, long-duration videos, image-to-text encoding