Shot boundary detection on broadcast news videos using a combination of multiple features

Shot boundary detection (SBD) is essential for video structural analysis, indexing, and retrieval. A video shot is a sequence of related images captured continuously by a single camera. Existing SBD approaches can generally be categorized into threshold-based, machine learning-based, and deep learning-based methods, which are often sensitive to variations in both video content and editing styles. This article presents a shot boundary detection method for broadcast news videos that combines modified block-based histogram features, speeded-up robust features (SURF) keypoint matching score features, and region covariance descriptors within an adaptive thresholding framework. The integration of object-level correspondence through SURF and covariance analysis enhances robustness against illumination variation and motion, effectively reducing false detections. The method operates on overlapping video segments to ensure consistent boundary localization. Experiments on 22 video sequences from news, broadcast videos, and TRECVID2001 data sets achieved an average precision of 0.91, recall of 0.93, and F1-score of 0.92. Statistical validation confirmed the significance of the improvements, demonstrating that the proposed method outperforms conventional handcrafted approaches and achieves comparable performance to recent deep learning-based methods. These results indicate that the method provides a reliable, generalizable performance suitable for broadcast news video segmentation, video archiving, and real-world media analysis.