Unsupervised Saliency Detection via kNN Mechanism and Object-Biased Prior.

Autor:	Zhou, Xin, Ren, Zhaohui, Zhou, Shihua, Yu, Tianzhuang, Jiang, Zeyu
Předmět:	SUPERVISED learning K-nearest neighbor classification K-means clustering RESEARCH personnel PRIOR learning
Zdroj:	Neural Processing Letters; Dec2023, Vol. 55 Issue 6, p8385-8399, 15p
Abstrakt:	In recent years, some researchers have put forth the compactness hypothesis, which suggests that similar colours tend to accumulate in the salient region as opposed to the non-salient region in image saliency detection. We discovered that the k-nearest neighbour (kNN) mechanism assumes the presence of similar objects in close proximity. As the kNN method is a supervised learning approach, we introduced an unfixed k value and combined it with the clustering idea of k-means to develop a novel algorithm called kNN clustering. We proposed an object-biased prior and an improved boundary and background prior based on a given compact matrix. Our algorithm was extensively tested on five publicly available datasets. The experimental results demonstrated that it outperformed eight existing top unsupervised models in producing high-quality saliency maps at full resolution. The improved prior methods were particularly effective when employed with existing algorithms lacking prior knowledge, especially with low-performing models. [ABSTRACT FROM AUTHOR]
Databáze:	Complementary Index
Externí odkaz:	Zobrazit plný text záznamu Full text from SpringerLink