An Integrated Dynamic Scene Algorithm for Segmentation and Motion Estimation
- PDF / 1,632,522 Bytes
- 9 Pages / 600 x 792 pts Page_size
- 89 Downloads / 268 Views
An Integrated Dynamic Scene Algorithm for Segmentation and Motion Estimation Ikhlas Abdel-Qader Electrical and Computer Engineering Department, College of Engineering and Applied Sciences, Western Michigan University, Kalamazoo, MI 49008-5314, USA Email: [email protected]
Tomislav Bujanovic Electrical and Computer Engineering Department, College of Engineering and Applied Sciences, Western Michigan University, Kalamazoo, MI 49008-5314, USA Email: [email protected] Received 8 June 2004; Revised 3 December 2004; Recommended for Publication by Luciano da F. Costa Segmentation and motion estimation are two problems that require accurate estimation for many applications in computer vision and image analysis. This work presents a solution to these two problems simultaneously. Both the segmentation and motion fields are integrated and estimated in parallel to reduce computation time. The presented algorithm is based on producing motion estimates and restored pixel intensity values through an optimization process that uses deterministic mean-field annealing (MFA) framework. The MFA results at different temperature values are used to run a segmentation process using the concept of region-growing-based algorithm. The segmentation process starts at high temperatures and continues in parallel to the annealing process to refine the segmentation process at lower temperatures. The algorithm results are good and dependent on the annealing parameters. Several experimental results from synthetic and real-world sequences are presented. Keywords and phrases: segmentation, motion estimation, mean-field annealing.
1.
INTRODUCTION
Accurate estimation of motion information and scene segmentation is the focus of investigators in many discipline areas for a variety of applications. Visual motion analysis is necessary for applications such as target tracking, video coding, automatic surveillance, remote sensing, image compression, and many other real-life applications. The estimation of the motion (displacement) fields is the first step in many applications. The methods of estimating the motion fields can be categorized into three groups: the gradient methods (known also as the optical flow); the correspondence methods; and block-matching methods. Each group has its own advantages, disadvantages, and limitations. In this work, the algorithm incorporates the methods of the first category. One of the main advantages of this category is the ability to provide dense displacement fields at subpixel accuracy as opposed to the algorithms in the other two categories that produce displacement vectors for blocks or for predetermined tokens (sparse motion fields) [1, 2]. Segmentation on the other hand aims to segment the scene into different objects or into objects and background. In general terms, image segmentation can be described as the process of generating
pixel labels at each pixel. These labels are intended to group pixels into different segments, objects, or partitions. Algorithms can be categorized as spatial segmentation, temporal se
Data Loading...
