Towards a Simple and Efficient Object-based Superpixel Delineation Framework

Abstract

Superpixel segmentation methods are widely used in computer vision applications due to their properties in border delineation. These methods do not usually take into account any prior object information. Although there are a few exceptions, such methods significantly rely on the quality of the object information provided and present high computational cost in most practical cases. Inspired by such approaches, we propose Object-based Dynamic and Iterative Spanning Forest (ODISF), a novel object-based superpixel segmentation framework to effectively exploit prior object information while being robust to the quality of that information. ODISF consists of three independent steps: (i) seed oversampling; (ii) dynamic path-based superpixel generation; and (iii) object-based seed removal. After (i), steps (ii) and (iii) are repeated until the desired number of superpixels is finally reached. Experimental results show that ODISF can surpass state-of-the-art methods according to several metrics, while being significantly faster than its object-based counterparts.

Efficient Multiscale Object-based Superpixel Framework

“…That is, considering different objects with distinct features, our approach's strategy of oversampling and objectbased seed removal is proven more effective than the typical approach of seed recomputation seen in SLIC, LSC and OISF. When analyzing the UE, the results support the findings of [5,9] claiming that object-based strategies offer a better superpixel leaking prevention.…”

“…In OISF, the user controls the superpixel displacement and morphology concerning the map's estimation. While it is slow and it is highly dependable on the map quality (i.e., propagates the maps' errors), Objectbased DISF (ODISF) [5] offers a highly accurate and faster solution with minimum influence of saliency errors.…”

“…Given that it is burdensome to evaluate all possible SICLE configurations, we elected the following one as starting point for our analysis based on previous works. We sampled using GRID with N 0 = 8000 seeds and considered w 1 as in [8], and used the V obj , considering V 4 , as presented in [5]. Finally, similarly to most seed-based methods [1,18,33], we set Ω = 10.…”

“…Figure 3 shows the comparison between SICLE and OISF, which uses such object-based path-cost function. Therefore, in this work, we consider the f max function due to its reportedly accurate performance in several works [8,5,10].…”

“…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.…”