Deep Decoupling of Defocus and Motion Blur for Dynamic Segmentation
We address the challenging problem of segmenting dynamic objects given a single space-variantly blurred image of a 3D scene captured using a hand-held camera. The blur induced at a particular pixel on a moving object is due to the combined effects of came
- PDF / 2,426,553 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 55 Downloads / 250 Views
Abstract. We address the challenging problem of segmenting dynamic objects given a single space-variantly blurred image of a 3D scene captured using a hand-held camera. The blur induced at a particular pixel on a moving object is due to the combined effects of camera motion, the object’s own independent motion during exposure, its relative depth in the scene, and defocusing due to lens settings. We develop a deep convolutional neural network (CNN) to predict the probabilistic distribution of the composite kernel which is the convolution of motion blur and defocus kernels at each pixel. Based on the defocus component, we segment the image into different depth layers. We then judiciously exploit the motion component present in the composite kernels to automatically segment dynamic objects at each depth layer. Jointly handling defocus and motion blur enables us to resolve depth-motion ambiguity which has been a major limitation of the existing segmentation algorithms. Experimental evaluations on synthetic and real data reveal that our method significantly outperforms contemporary techniques.
Keywords: Segmentation blur
1
·
Neural network
·
Defocus blur
·
Motion
Introduction
Segmentation of dynamic objects in a scene is a widely researched problem as it forms the first step for many image processing and computer vision applications such as surveillance, action recognition, scene understanding etc. Classical video-based motion segmentation algorithms [8,22] assume that the camera is stationary and only the object of interest is in motion in the scene, thus allowing them to learn the static background and separate out the dynamic object. However, the assumption of a static camera does not hold in most real-world applications – the camera might be hand-held or mounted on a moving vehicle Electronic supplementary material The online version of this chapter (doi:10. 1007/978-3-319-46478-7 46) contains supplementary material, which is available to authorized users. c Springer International Publishing AG 2016 B. Leibe et al. (Eds.): ECCV 2016, Part VII, LNCS 9911, pp. 750–765, 2016. DOI: 10.1007/978-3-319-46478-7 46
Deep Decoupling of Defocus and Motion Blur for Dynamic Segmentation
(a)
(b)
(c)
751
(d)
Fig. 1. Dynamic scenes. (a and b) Blur perception dataset [23], (c) a frame extracted from a video downloaded from the internet, and (d) an indoor image we captured ourselves using a hand-held camera.
and there could be significant parallax effects due to the 3D nature of the scene. The combination of a moving camera and a dynamic 3D scene often introduces an additional challenge in the form of blurring. To bring the entire 3D scene into focus, one must select a small aperture (large depth of field), and thereby a larger exposure time. But this increases the chances of motion blur since both object and camera are in motion. On the other hand, reducing the exposure time by choosing a large aperture (small depth of field) results in depth dependent defocus blur. Thus, there exists a trade-off between defocus and motio
Data Loading...
