Recently, the increased use of online video has been driven by the accessibility of sophisticated recording devices, leading to a huge expansion of video datasets. These datasets provide valuable visual information for affective computing applications like personality trait recognition (PTR). However, processing huge numbers of video frames presents significant computing challenges, particularly in terms of storage and memory utilization. Thus, we address this issue by proposing a key frame selection method that selects the most informative frames representing the video content.
The first step in our method is to find candidate frames by detecting significant changes between consecutive frames. After that, grayscale conversion, feature extraction, and hierarchical density-based spatial clustering of applications clustering are used to group similar frames. Finally, the best frame from each cluster is selected based on brightness for visibility and sharpness measured by the Laplacian score. This key frame selection method significantly reduces computational costs compared to the widely used random frame selection approach in previous PTR studies.
The experimental results on the ChaLearn dataset show a reduction of about 86% in frame extraction using the key frame method compared to the random frame method while maintaining a competitive prediction accuracy of around 89%. By using key frame selection, the method improves the efficiency of extracting meaningful and relevant frames from video. It also reduces frame redundancy and lowers computational costs. Its competitive prediction accuracy makes PTR with key frame selection a promising method for real-time applications such as automated personality assessment in hiring processes, psychological assessments and mental health analysis.
This study presents a novel key frame selection approach for PTR, utilizing clustering and sharpness-based selection to enhance efficiency. The method significantly reduces computational costs by 86% while maintaining 89% accuracy, demonstrating its potential for real-time applications in automated personality assessment, recruitment and mental health analysis.
