| Autor: |
Devi, J. Sarala, Sulthana, A. Razia |
| Předmět: |
|
| Zdroj: |
Multimedia Tools & Applications; Feb2023, Vol. 82 Issue 5, p6769-6785, 17p |
| Abstrakt: |
Video Object Segmentation (VOS) for separating a foreground object from a video sequence is an intricate task and relies on fine-tuning. Many recent approaches focus on pixel-wise matching of foreground objects and gives importance to balancing the relation between the pixels for identifying the foreground objects and might lead to misclassification. This paper explores mapping between the foreground and background objects in semi-supervised VOS by balancing and mutually mapping the pixels between the foreground and background objects. The proposed model makes practical and effective use of enhanced pixel and instance level matching to improve the prediction. Moreover, the framework implements ensemble learning with a Leaky-ReLU activation function that improves the segmentation process. To evaluate the results of object segmentation process, J and F scores are measured. We carry experiments broadly on popular benchmark DAVIS, in the versions 2016 and 2017. Our Model achieves a promising performance of J & F score of 82%, surpassing all the other techniques. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink