Automatic trailer generation for movies using convolutional neural network.

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Automated movie trailer generation is challenging due to the need for accurate genre identification and compelling scene selection.
• Existing methods often lack the precision required for high-quality trailer production.

Purpose of the Study:

• To propose a novel, two-stage method for automated movie trailer production.
• To enhance genre identification accuracy and trailer production quality using deep learning and text analysis.

Main Methods:

• Genre identification using convolutional neural networks (CNNs) on poster images and TF-IDF vectors from subtitles.
• Classification of movie genres using a classification and regression tree.
• Genre-specific CNN models trained on key sequences for trailer production, analyzing visual and textual scene features.

Main Results:

• Achieved 83.39% accuracy in movie genre detection, surpassing existing methods by at least 1%.
• Demonstrated 56.69% precision in trailer generation, an improvement of at least 8% over comparative techniques.

Conclusions:

• The proposed method effectively automates movie trailer production through accurate genre identification and intelligent scene selection.
• This approach offers significant improvements in both the accuracy of genre recognition and the quality of generated trailers.