Superpixel Generation by the Iterative Spanning Forest Using Object Information

Abstract

Superpixel segmentation methods aim to partition the image into homogeneous connected regions of pixels (i.e., superpixels) such that the union of its comprising superpixels precisely defines the objects of interest. However, the homogeneity criterion is often based solely on color, which, in certain conditions, might be insufficient for inferring the extension of the objects (e.g., low gradient regions). In this dissertation, we address such issue by incorporating prior object information — represented as monochromatic object saliency maps — into a state-of-the-art method, the Iterative Spanning Forest (ISF) framework, resulting in a novel framework named Object-based ISF (OISF). For a given saliency map, OISF-based methods are capable of increasing the superpixel resolution within the objects of interest, whilst permitting a higher adherence to the map’s borders, when color is insufficient for delineation. We compared our work with state-of-the-art methods, considering two classic superpixel segmentation metrics, in three datasets. Experimental results show that our approach presents effective object delineation with a significantly lower number of superpixels than the baselines, especially in terms of preventing superpixel leaking.

Towards a Simple and Efficient Object-based Superpixel Delineation Framework

“…Given such information, these algorithms may improve the delineation performance in relevant regions. The Object-based ISF [10], [15] method is a three-stage IFT-based superpixel approach that falls into this category. In OISF, object information is represented by an object saliency map.…”

Efficient Multiscale Object-based Superpixel Framework

“…Saliency-based Superpixels (SS) [37] can be classified as such, even though such term was first coined in [6]. Recent object-based solutions can be seen as generalizations of previous methods for handling with saliency maps, such as Object-based ISF (OISF) [6,7]. In OISF, the user controls the superpixel displacement and morphology concerning the map's estimation.…”

“…We selected several state-of-the-art superpixel methods based on their effectiveness and efficiency: (i) SLIC [1] 1 ; (ii) LSC [18] 2 ; (iii) SH [34] 3 ; (iv) ERS [20] 4 ; (v) OISF-OSMOX [7] 5 ; and (vi) SEAL-ERS [32] 6 . For all methods, the recommended parameter configuration was set.…”

“…We emphasize that the GT maps are only used in this particular analysis. It is important to note that, differently from other object-based strategies [6,7], the delineation performance of SICLE is not impacted by the quality of the map's border. As consequence, and setting aside that generating ideal maps is yet an unfeasible task, we claim that such property is positive since it disregards the need to improve the object saliency map whenever it is not the problem's main goal.…”

“…These object-based methods use such information for favoring the delineation of the object of interest at the expense of others, resulting in a more accurate superpixel segmentation. However, exhaustively exploiting saliency information can lead to higher computational cost and saliency quality dependency, as seen in Objectbased Iterative Spanning Forest (OISF) [6,7] variants. Conversely, Object-based DISF (ODISF) [5] overcomes the latter drawback while still being slightly slower than object-unaware methods.…”