Graph-based superpixel labeling for enhancement of online video segmentation

Abstract

In this paper, we propose a novel approach for video segmentation. The proposed work is based on exploiting a superpixel-based image segmentation approach to improve the performance of state-of-the-art foreground/background segmentation techniques. A fusion between a bilayer segmentation and a geodesic segmentation approaches with a graph-based superpixel segmentation method is performed. Four different combination alternatives are investigated in terms of performance and efficiency. Manually-labeled ground truth video sequences as well as our own recorded video sequences were used for evaluation purposes. The evaluation results confirm the potential of the proposed method in enhancing the accuracy of the video segmentation over the state-of-the-art.

A Model of Pixel and Superpixel Clustering for Object Detection

“…With increasing image resolution, the subject matter of superpixels becomes especially relevant due to the necessity to reduce the computational complexity of image processing by replacing pixel operations with superpixel ones. Therefore, it is widely discussed in the literature, albeit in a non-strict, heuristic formulation of the problem of calculating superpixels as some enlarged pixels, which are usually identified with connected image segments that do not distort the boundaries of visually observed objects [ 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 ]. There are no generally accepted accurate definitions for superpixels.…”

“…Various approaches to the calculation of superpixels are considered in detail in [ 29 ]. A stable stereotype is the formation of superpixels in the mode of local pixel enlargement [ 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 ], although for color images there are also heuristic divisive approaches to determining superpixels [ 37 ]. When the pixels are enlarged, the boundaries between the objects are masked by the textural features of the reflecting surfaces of the objects in the scene, which are actually used to select the objects-of-interest in the images of a given subject area.…”

“…The practical calculation of superpixels for individual images is useful to find suitable heuristic criteria for pixel merging in high-speed agglomerative pixel clustering methods that are being developed to solve current engineering problems of detection and recognition of objects-of-interest. Experimental verification of the effectiveness of a priori assumptions about the unification of superpixels in terms of the number of pixels, size, shape, and other features, as in [ 27 , 28 , 29 , 30 , 31 , 32 , 33 , 34 , 35 , 36 , 37 ], remains an important research area to refine the concept of superpixels and optimize the calculations, but only to the extent that it does not reduce the efficiency of approximation error minimization.…”