The Human Image Segmentation Algorithm Based on Face Detection and Biased Normalized Cuts
Attributed to pose variation (frontal, profile, et.), the color and texture difference of clothes, the presence of noise, low contrast, uneven illumination and complex background. There are enormous difficultly in human image segmentation. In this paper,
- PDF / 1,223,592 Bytes
- 10 Pages / 439.37 x 666.142 pts Page_size
- 28 Downloads / 161 Views
Introduction
Human body segmentation in human images is a very important step in many computer vision tasks, such as image processing, video tracking, pose estimation, content-based image retrieval, pedestrian detection, action understanding, etc. However, to segment a human body in a human image is still a very challenging task because of segmentation is inherently ill-posed, the appearance and pose variation, the presence of noise, low contrast, and intensity inhomogeneity. In the last decade the most popular approach to interacitve image segmentaiton in computer vision was graph cut. To avoid the minimum cut criteria favors cutting small sets of isolated nodes in the graph. Using the volume for c Springer-Verlag Berlin Heidelberg 2015 H. Zha et al. (Eds.): CCCV 2015, Part I, CCIS 546, pp. 134–143, 2015. DOI: 10.1007/978-3-662-48558-3 14
The Human Image Segmentation Algorithm Based on Face Detection
135
the normalized weights. It aims at extracting the global impression of an image. The normalized cuts [1] criterion measures both the total dissimilarity between the different groups as well as the total similarity within the groups. Subhransu Maji [2] present a modification of ”Normalized cuts” to incorporate priors which can be used for constrained image segmentation. In this paper, we employ face detection and biaed normalized cuts to segment human body in static image. Different from the previous methods, our approach requires much less training data for face detection, and seeds estimation model is simple and effective. Moreover, our method is different from biased normalized cuts which we have better constraints and need to do a region merging after biaed normalized cuts segmentation. Also, our method does not require human interaction, and it is a fully automated segmentation method. Our segmentation results are more accurate and effective. The rest of this paper is organized as follows. Section 2 discusses the most related work with ours. Section 3 describes the details of our proposed method. Analysis and experimental results in Section 4. Finally, Section 5 concludes the paper and propose some future work..
2 2.1
Related Work Face Detection
Face detection is dominated by discriminatively-trained scanning window classifiers [3], most ubiquitous of which is the Viola Jones detector [4]. Zhu [5] model was based on a mixtures of trees with a shared pool of parts. They modeled every facial landmark as a part and used global mixtures to capture topological changes due to viewpoint. Their system was also trained discriminatively, but with much less training data, particularly when compared to commercial systems. 2.2
Human Image Segmentation
Ashwini T. Magar et. [6] divided human segmentation techniques to exemplar based, part based and other based. In exemplar based approach [7–9], an exemplar pool should be constructed first, and then, the test images was matched with the exemplars. Agarwal and Triggs [10] modeled the image window with a dense grid of local gradient orientation histograms to select similar
Data Loading...
