Efficient Video Object Segmentation Using JIT K-Means Clustering

Video segmentation dissects video sequences into distinct regions based on visual cues like object borders, movement patterns, color variations, and textures. This partitioning aims to isolate individual objects and temporal events from the background, ultimately revealing a more structured and interpretable representation of the video content. In this work, we propose K-means clustering and MATLAB Just-In-Time (JIT) accelerator within a video segmentation framework to boost both accuracy and efficiency, especially in resource-limited environments. This work aims to evaluate K-means for video object segmentation, assess JIT's impact on its speed and resource usage, and optimize their integration considering feature representation and clustering parameters. Our experimental results show that the MATLAB JIT is able to achieve a significant speedup compared to the video segmentation in a normal mode with the average acceleration = 89.88%.