Walsh-Hadamard transform kernel-based feature vector for shot boundary detection.

Area of Science:

• Computer Vision
• Digital Signal Processing
• Multimedia Analysis

Background:

• Video Shot Boundary Detection (SBD) is crucial for video analysis tasks like summarization and retrieval.
• Existing SBD methods often struggle with accuracy, especially for gradual transitions.

Purpose of the Study:

• To propose a novel and robust Video Shot Boundary Detection (SBD) method.
• To improve the accuracy of detecting both abrupt and gradual shot transitions in videos.

Main Methods:

• Features including color, edge, texture, and motion strength are extracted using Walsh-Hadamard Transform (WHT).
• Feature significance is used to calculate weights, forming a continuity signal for shot transition identification (PBI).
• The method was evaluated using TRECVID 2007 and Openvideo datasets, and compared with existing SBD methods.

Main Results:

• The proposed method achieved an F1-Score of 97.4% for cut transitions, 78% for gradual transitions, and 96.1% overall.
• WHT-based features demonstrated superior performance compared to other existing methods when evaluated with a support vector machine classifier.
• The system evaluation confirmed the robustness of the proposed SBD method.

Conclusions:

• The proposed WHT-based feature extraction method offers a significant improvement in video shot boundary detection.
• The approach is effective in accurately classifying both abrupt and gradual shot transitions.
• This method provides a robust and efficient solution for video analysis and content retrieval.