Fig 2 - uploaded by Haihua Feng
Content may be subject to copyright.
(a) True synthetic LADAR range image and the initial curve (t = 0). (b) Noisy range image (P = 5%; = 10; 1R = 1000). (c) Reconstructed range image using CE-EM algorithm and the final estimated target boundary (t = 156). (d) Reconstructed range image using MR-EM algorithm (8 2 8 pixel block).
Source publication
In this paper, we develop a new unified approach for laser radar range anomaly suppression, range profiling, and segmentation. This approach combines an object-based hybrid scene model for representing the range distribution of the field and a statistical mixture model for the range data measurement noise. The image segmentation problem is formulat...
Contexts in source publication
Context 1
... this example, we use the synthetic LADAR image shown in Fig. 2(a), which consists of 256 256 pixels. The true back- ground can be represented by a polynomial ...
Context 2
... scalar weights and in (4) were selected empirically to yield the visually best re- constructions. The results were found relatively insensitive to variation of these parameters within a range of %. 1 The probability of anomaly P is determined by the CNR of radar and the ratio between the range uncertainty interval 1R and the range resolution [5]. Fig. 2(a) shows the true range image. The initial curve, which is a small circle, is also shown in Fig. 2(a). The initial curve serves as the starting point of our CE-EM algorithm. Fig. 2(b) shows the raw range data. After iterations, the curve converges to the true target boundary as shown in Fig. 2(c), and the corresponding target and ...
Context 3
... results were found relatively insensitive to variation of these parameters within a range of %. 1 The probability of anomaly P is determined by the CNR of radar and the ratio between the range uncertainty interval 1R and the range resolution [5]. Fig. 2(a) shows the true range image. The initial curve, which is a small circle, is also shown in Fig. 2(a). The initial curve serves as the starting point of our CE-EM algorithm. Fig. 2(b) shows the raw range data. After iterations, the curve converges to the true target boundary as shown in Fig. 2(c), and the corresponding target and background range value estimates are also given in Fig. 2(c). Note that the fine features such as the ...
Context 4
... a range of %. 1 The probability of anomaly P is determined by the CNR of radar and the ratio between the range uncertainty interval 1R and the range resolution [5]. Fig. 2(a) shows the true range image. The initial curve, which is a small circle, is also shown in Fig. 2(a). The initial curve serves as the starting point of our CE-EM algorithm. Fig. 2(b) shows the raw range data. After iterations, the curve converges to the true target boundary as shown in Fig. 2(c), and the corresponding target and background range value estimates are also given in Fig. 2(c). Note that the fine features such as the object tip and corners are well preserved in the reconstructed image and the estimated ...
Context 5
... uncertainty interval 1R and the range resolution [5]. Fig. 2(a) shows the true range image. The initial curve, which is a small circle, is also shown in Fig. 2(a). The initial curve serves as the starting point of our CE-EM algorithm. Fig. 2(b) shows the raw range data. After iterations, the curve converges to the true target boundary as shown in Fig. 2(c), and the corresponding target and background range value estimates are also given in Fig. 2(c). Note that the fine features such as the object tip and corners are well preserved in the reconstructed image and the estimated ...
Context 6
... initial curve, which is a small circle, is also shown in Fig. 2(a). The initial curve serves as the starting point of our CE-EM algorithm. Fig. 2(b) shows the raw range data. After iterations, the curve converges to the true target boundary as shown in Fig. 2(c), and the corresponding target and background range value estimates are also given in Fig. 2(c). Note that the fine features such as the object tip and corners are well preserved in the reconstructed image and the estimated ...
Context 7
... uses an automated coarse-to-fine termination procedure based on a statistical cri- teria which adjusts resolution so that the estimated number of anomalous pixels at the chosen resolution is consistent with the anomaly probability (which is assumed known). For this ex- ample with %, the optimal resolution according to [11] is 8 8 pixel blocks. In Fig. 2(d), we show the MR-EM recon- structed range image at the corresponding optimal resolution of 8 8 pixel blocks. From Fig. 2(d), we can see that while the re- construction suppresses all anomalous pixels, it fails to preserve some fine object features such as tips and corners of the ...
Context 8
... estimated number of anomalous pixels at the chosen resolution is consistent with the anomaly probability (which is assumed known). For this ex- ample with %, the optimal resolution according to [11] is 8 8 pixel blocks. In Fig. 2(d), we show the MR-EM recon- structed range image at the corresponding optimal resolution of 8 8 pixel blocks. From Fig. 2(d), we can see that while the re- construction suppresses all anomalous pixels, it fails to preserve some fine object features such as tips and corners of the ...
Context 9
... we show the estimated target boundary of our new CE-EM method indicated as the red curve and the true target boundary is indicated as the blue curve. The target boundary and scene ranges have been well recovered. In Fig. 4(b), we show the corresponding ad hoc target boundary estimation result obtained by applying (13) to the MR-EM results in Fig. 2(d). The estimated boundary does not recover the target tips and edges well. The blocking introduced by the MR-EM anomaly suppression method (necessary to obtain optimal anomaly suppression within a planar profiling con- text), serves as a confounding factor to subsequent boundary ...
Context 10
... further illustrate the impact of the anomaly probability on the reconstruction results, we use the same synthesized range image in Fig. 2 but increase the anomaly probability to 12%. Fig. 3(a) is the true range image and the initial curve. Fig. 3(b) shows the noisy range data. The curve converges to the target boundary as shown in Fig. 3(c), and a good estimate of the range values is also given in Fig. 3(c). For comparison, the result using the MR-EM approach of [11] are ...
Similar publications
The colors present in an image of a scene provide information about its constituent elements. But the amount of information depends on the imaging conditions and on how information is calculated. This work had two aims. The first was to derive explicitly estimators of the information available and the information retrieved from the color values at...
Implicit representations of curves have certain advantages over explicit representation, one of them being the ability to determine with ease whether a point is inside or outside the curve (inside-outside functions). However, save for some special cases, it is not known how to construct implicit representations which are guaranteed to preserve the...
![Figure 1. This refinement of a previously published diagram [8], shows...](profile/Lewis-Griffin/publication/6937058/figure/fig1/AS:667210343276548@1536086750311/This-refinement-of-a-previously-published-diagram-8-shows-the-psychological-structure_Q320.jpg)
Categorization of colour has been widely studied as a window into human language and cognition, and quite separately has been used pragmatically in image-database retrieval systems. This suggests the hypothesis that the best category system for pragmatic purposes coincides with human categories (i.e. the basic colours). We have tested this hypothes...
Many classical image denoising methods are based on a local averaging of the color, which increases the signal/noise ratio. One of the most used algorithms is the neighborhood filter by Yaroslavsky or sigma filter by Lee, also called in a variant "SUSAN" by Smith and Brady or "Bilateral filter" by Tomasi and Manduchi. These filters replace the actu...
A novel, physically motivated deformable model for shape recovery and segmentation is presented. The model, referred to as the charged-particle model (CPM), is inspired by classical electrodynamics and is based on a simulation of charged particles moving in an electrostatic field. The charges are attracted towards the contours of the objects of int...
Citations
... Developed by Dempster et al. [6], the expectation maximization (EM) algorithm is a general method of finding the ML estimate of the parameters of an underlying distribution from the given observed data set when the data is incomplete or has missing values. It is particularly useful in situations where optimizing the likelihood function is analytically intractable, but the likelihood function can be simplified by assuming the existence of missing data [7, 8]. In our proposed method, the true target states and the measurementto-target association is treated as the missing data. ...
In this paper, we propose a maximum likelihood (ML) approach to address the joint registration, association and fusion problem in multi-sensor and multi-target surveillance. In particular, an expectation maximization (EM) algorithm is employed here. At each iteration of the EM, the extended Kalman filter (EKF) is incorporated into the E-step to obtain the fusion results, while the registration parameters are updated in the M-step. Association of sensor measurements to the targets are also computed as the missing data in the E-step. The main advantage of the proposed method compared to the conventional approaches is that the mutual effects of registration, association and fusion are taken into the consideration when formulating the multi-sensor, multi-target tracking problem. The simulation results demonstrate that the performance of the proposed method in terms of mean square error (MSE) is close to the posterior Cramer-Rao bound (PCRB), and is better than one of the conventional approaches that perform registration, association and fusion separately.
This study is to assess the optimal process for the surface smoothing of 3D image model of in vivo human inner ear. A 3D image model of the inner ear including semicircular canals, vestibule and cochlea was reconstructed from magnetic resonance imaging data. Three smoothing methods
including constrained smoothing, unconstrained smoothing and smoothsurface will be discussed. The volume of the 3D image model produced by unconstrained smoothing differed substantially from the original model volume prior to smoothing. Constrained smoothing had an uneven effect on the surface
of the 3D image models. Using the smoothsurface module, we were able to obtain an optimal surface of the 3D image model of the human inner ear including the semicircular canals, vestibule and cochlea, using 10 iterations and a lambda value of 0.6. The 3D image model by smoothsurface would
be worthy of reference in clinical evaluation.
To perform surveillance using multiple cameras, camera calibration, measurement-to-object association, fusion of measurements from multiple cameras are three essential components. While these three issues are usually addressed separately, they actually have mutual effects on each other. For example, calibration requires correctly associated objects and measurements with calibration errors will result in wrong associations. In this paper, we present a novel joint calibration, association and fusion approach for multi-camera tracking. More specifically, the expectation-maximization (EM) algorithm is incorporated with the extended Kalman filter (EKF) to give a simultaneous estimate of object states, calibration and association parameters. The real video data collected from two cameras are used to evaluate the tracking performance of the proposed method. Compared to the conventional methods, which perform calibration, association and fusion separately, it is shown that the proposed method can significantly improve the robustness and the accuracy of multi-object tracking.
