DBTSF-VSOD: a decision-based two-stage framework for video salient object detection

• Published in: International journal of multimedia information retrieval 2024-12, Vol.13 (4), p.38, Article 38
• Main Authors: Kumain, Sandeep Chand, Singh, Maheep, Awasthi, Lalit Kumar
• Format: Article
• Language: English
• Subjects: Computer Science; Computer vision; Content analysis; Data Mining and Knowledge Discovery; Database Management; Datasets; Decision making; Deep learning; Digital computers; Digital imaging; Flow mapping; Image compression; Image processing; Image Processing and Computer Vision; Information Storage and Retrieval; Information Systems Applications (incl.Internet); Methods; Multimedia Information Systems; Object recognition; Optical flow (image analysis); Regular Paper; Salience; Video compression; Video data
• ISSN: 2192-6611; 2192-662X
• DOI: 10.1007/s13735-024-00346-4

Salient Object Detection (SOD) plays a pivotal role in digital image processing, focusing on identifying objects in images or videos that naturally draw human attention. In the world of computer vision and image processing, these eye-catching objects are known as salient objects. The process of saliency detection holds significant importance in various computer vision applications, including video compression, automated cropping, and video summarization. Recent advancements in AI technology have witnessed the emergence of numerous deep learning-based end-to-end networks in the video saliency domain. However, these approaches predominantly concentrate on network architecture design. In this study, the author(s) introduce a decision-based two-stage framework for video saliency detection, in which the decision is taken based on frame content analysis. This approach enhances the quality of the saliency map in the overall process. In the proposed model, the process unfolds in two stages. Initially, an initial saliency map is generated. Subsequently, in the second phase, the final saliency output is produced by leveraging both the initial saliency map and an optical flow map. To evaluate the effectiveness of the proposed model extensive assessments have been carried out on widely recognized public datasets namely VOS, DAVSOD, and ViSAL. The experimental findings indicate that the proposed model delivers strong performance, exhibiting competitiveness in comparison to state-of-the-art methods on the metrics S-measure, F-measure, and MAE.