Superpixel Analysis for Object Detection and Tracking with Application to UAV Imagery

Rasmussen, Christopher

doi:10.1007/978-3-540-76858-6_5

Cited by 8 publications

(2 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Superpixel segmentation is an important pre-processing step of many image processing algorithms, like object detection and tracking (Rasmussen, 2007), image segmentation and modelling (Mi et al, 2009), salience detection (Tong et al, 2014) and image and object classification (Liu et al, 2016), etc. A superpixel is a set of pixels which are homogeneous in perception or spectrum, a representative color or spectrum is used to represent each superpixel, and the effect of noise and distortions is thus mitigated.…”

Section: Introductionmentioning

confidence: 99%

Object Manifold Alignment for Multi-Temporal High Resolution Remote Sensing Images Classification

Gao¹,

Zhang²,

Gu³

2017

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

ABSTRACT:Multi-temporal remote sensing images classification is very useful for monitoring the land cover changes. Traditional approaches in this field mainly face to limited labelled samples and spectral drift of image information. With spatial resolution improvement, "pepper and salt" appears and classification results will be effected when the pixelwise classification algorithms are applied to highresolution satellite images, in which the spatial relationship among the pixels is ignored. For classifying the multi-temporal high resolution images with limited labelled samples, spectral drift and "pepper and salt" problem, an object-based manifold alignment method is proposed. Firstly, multi-temporal multispectral images are cut to superpixels by simple linear iterative clustering (SLIC) respectively. Secondly, some features obtained from superpixels are formed as vector. Thirdly, a majority voting manifold alignment method aiming at solving high resolution problem is proposed and mapping the vector data to alignment space. At last, all the data in the alignment space are classified by using KNN method. Multi-temporal images from different areas or the same area are both considered in this paper. In the experiments, 2 groups of multi-temporal HR images collected by China GF1 and GF2 satellites are used for performance evaluation. Experimental results indicate that the proposed method not only has significantly outperforms than traditional domain adaptation methods in classification accuracy, but also effectively overcome the problem of "pepper and salt".

show abstract

Section: Introductionmentioning

confidence: 99%

Object Manifold Alignment for Multi-Temporal High Resolution Remote Sensing Images Classification

Gao¹,

Zhang²,

Gu³

2017

Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci.

View full text Add to dashboard Cite

show abstract

“…Image over-segmentation has been widely applied in various computer vision pipelines, such as segmentation [37,36,11], recognition [13], tracking [28], localization [7] and modeling [10,26]. In these applications, over-segments (also known as superpixels in [29]) represent small regions with homogeneous appearance and conform to local image structure, and thus they provide a better support for region-based features than local windows.…”

Section: Introductionmentioning

confidence: 99%

Structure-Sensitive Superpixels via Geodesic Distance

et al. 2012

View full text Add to dashboard Cite

Over-segments (i.e. superpixels) have been commonly used as supporting regions for feature vectors and primitives to reduce computational complexity in various image analysis tasks. In this paper, we describe a structuresensitive over-segmentation technique by exploiting Lloyd's algorithm with a geodesic distance. It generates smaller superpixels to achieve lower under-segmentation in structuredense regions with high intensity or color variation, and produces larger segments to increase computational efficiency in structure-sparse regions with homogeneous appearance. We adopt geometric flows to compute the geodesic distances amongst pixels, and in the segmentation procedure, the density of over-segments is automatically adjusted according to an energy functional that embeds color homogeneity, structure density and compactness constraints. Comparative experiments with the Berkeley database show that the proposed algorithm outperforms prior arts while offering a comparable computational efficiency with fast methods, such as TurboPixels.

show abstract