An Enhanced Method For Evaluating Automatic Video Summaries

TL;DR

This paper introduces a new evaluation method for automatic video summaries that combines color and texture features with Bhattacharya distance, providing more perceptual and accurate quality assessments compared to single-feature approaches.

Contribution

The paper presents a novel evaluation approach that integrates multiple features and a specific dissimilarity measure, improving the accuracy of automatic video summary assessments.

Findings

The proposed method outperforms existing single-feature evaluation techniques.

It provides a more perceptual and reliable measure of summary quality.

Experimental results validate the effectiveness of combining color and texture features.

Abstract

Evaluation of automatic video summaries is a challenging problem. In the past years, some evaluation methods are presented that utilize only a single feature like color feature to detect similarity between automatic video summaries and ground-truth user summaries. One of the drawbacks of using a single feature is that sometimes it gives a false similarity detection which makes the assessment of the quality of the generated video summary less perceptual and not accurate. In this paper, a novel method for evaluating automatic video summaries is presented. This method is based on comparing automatic video summaries generated by video summarization techniques with ground-truth user summaries. The objective of this evaluation method is to quantify the quality of video summaries, and allow comparing different video summarization techniques utilizing both color and texture features of the video frames and using the Bhattacharya distance as a dissimilarity measure due to its advantages. Our Experiments show that the proposed evaluation method overcomes the drawbacks of other methods and gives a more perceptual evaluation of the quality of the automatic video summaries.