Objects In Clutter Research Articles

Object Detection by Spectropolarimeteric Imagery Fusion

In the past few years, imaging spectroscopy has been widely used. However, it only acquires intensity information in a narrow electromagnetic band, ignoring the polarimetric information of the electromagnetic wave and resulting in inaccurate object detection. According to electromagnetic theory, the reflected spectral signature depends on the elemental composition of objects residing within the scene, and the radiation's polarization characteristic is sensitive to surface features, such as relative smoothness and conductance. Independently, spectral and polarimetric features give incomplete representations of an object of interest. These representations are complementary, and it is expected that the combination of complementary information will reduce false alarms, improve confidence in target identification, and improve the quality of the scene description. Imaging spectropolarimetric technology as a new sensing method can acquire polarimetric information at narrow electromagnetic bands, but there are a few results showing how to combine these complementary features to detect objects in clutter. In this paper, a spectropolarimetric projection scheme is proposed to divide the spectropolarimetric data set into two parts: (1) a polarimetric spectrum data set and (2) a polarimetric data cube. Then, the polarimetric spectrum anomaly feature extraction method is used to deal with the polarimetric spectrum data set, and the adaptive polarimetric information fusion method is proposed to extract the feature from the polarimetric data cube. Finally, these features are combined by the Choquet integral to achieve better detection performance. The algorithm is applied to one complex scene, and detailed detection performance is evaluated.

Visual Recognition and Inference Using Dynamic Overcomplete Sparse Learning

We present a hierarchical architecture and learning algorithm for visual recognition and other visual inference tasks such as imagination, reconstruction of occluded images, and expectation-driven segmentation. Using properties of biological vision for guidance, we posit a stochastic generative world model and from it develop a simplified world model (SWM) based on a tractable variational approximation that is designed to enforce sparse coding. Recent developments in computational methods for learning overcomplete representations (Lewicki & Sejnowski, 2000; Teh, Welling, Osindero, & Hinton, 2003) suggest that overcompleteness can be useful for visual tasks, and we use an overcomplete dictionary learning algorithm (Kreutz-Delgado, et al., 2003) as a preprocessing stage to produce accurate, sparse codings of images. Inference is performed by constructing a dynamic multilayer network with feedforward, feedback, and lateral connections, which is trained to approximate the SWM. Learning is done with a variant of the back-propagation-through-time algorithm, which encourages convergence to desired states within a fixed number of iterations. Vision tasks require large networks, and to make learning efficient, we take advantage of the sparsity of each layer to update only a small subset of elements in a large weight matrix at each iteration. Experiments on a set of rotated objects demonstrate various types of visual inference and show that increasing the degree of overcompleteness improves recognition performance in difficult scenes with occluded objects in clutter.

Object recognition in dense clutter

Observers in recognition experiments invariably view objects against a blank background, whereas observers of real scenes sometimes view objects against dense clutter. In this study, we examined whether an object's background affects the information used for recognition. Our stimuli consisted of color photographs of everyday objects. The photographs were organized either as a sparse array, as is typical of a visual search experiment, or as high density clutter, such as might be found in a toy chest, a handbag, or a kitchen drawer. The observer's task was to locate an animal, vehicle, or food target in the stimulus. We varied the information in the stimuli by convolving them with a low-pass filter (blur) or a high-pass filter (edge) or converting them to grayscale. In two experiments, we found that the blur and edge manipulations produced a modest decrement in performance with the sparse arrangement but a severe decrement in performance with the clutter arrangement. These results indicate that the information used for recognition depends on the object's background. Thus, models of recognition that have been developed for isolated objects may not generalize to objects in dense clutter.

Search for a category target in clutter.

An airport security worker searching a suitcase for a weapon is engaging in an especially difficult search task: the target is not well-specified, it is not salient, and it is not predicted by its context. Under these conditions, search may proceed item-by-item. In the experiment reported here we tested whether the items for this form of search are whole familiar objects. Our displays were composed of color photographs of ordinary objects, that were either uniform in color and texture (simple), or had two or more parts with different colors or textures (compound). The observer's task was to detect the presence of a target belonging to a broad category (food). We found that when the objects were presented in a sparse array, search times to find the target were similar for displays composed of simple and compound objects. But when the same objects were presented as dense clutter, search functions were steeper for displays composed of compound objects. We attribute this difference to the difficulty of segmenting compound objects in clutter: compared with simple objects, compound objects are less likely to be organized into a single object by bottom--up grouping processes. Our results indicate that while search rates in a sparse display may be determined by the number of objects, search rates in clutter are also affected by the number of object parts.

Recognizing and segmenting objects in clutter

When viewing a cluttered scene, observers may not be able to segment whole objects prior to recognition. Instead, they may segment and recognize these objects in a piecemeal way. Here we test whether observers can use the appearance of one object part to predict the location and appearance of other object parts. During several training sessions, observers studied an object against a blank background. They then viewed this object against a background of clutter that camouflaged some parts of the object while leaving other parts salient. The observer’s task was to find the camouflaged part. We varied the symmetry of the salient part with the expectation that as this symmetry decreased, the information about the camouflaged part’s location and appearance would increase and this would facilitate search. Our results suggest that observers can use the salient part to predict the location, but not the appearance, of the camouflaged part.

Detection of small objects in clutter using a GA-RBF neural network

Detection of small objects in a radar or satellite image is an important problem with many applications. Due to a recent discovery that sea clutter, the electromagnetic wave backscatter from a sea surface, is chaotic rather than purely random, computational intelligence techniques such as neural networks have been applied to reconstruct the chaotic dynamic of sea clutter. The reconstructed sea clutter dynamical system which usually takes the form of a nonlinear predictor does not only provide a model of the sea scattering phenomenon, but it can also be used to detect the existence of small targets such as fishing boats and small fragments of icebergs by observing abrupt changes in the prediction error. We applied a genetic algorithm (GA) to obtain an optimal reconstruction of sea clutter dynamic based on a radial basis function (RBF) neural network. This GA-RBF uses a hybrid approach that employes a GA to search for the optimum values of the following RBF parameters: centers, variance, and number of hidden nodes, and uses the least square method to determine the weights. It is shown here that if the functional form of an unknown nonlinear dynamical system can be represented exactly using an RBF net (i.e., no approximation error), this GA-RBF approach can reconstruct the exact dynamic from its time series measurements. In addition to the improved accuracy in modeling sea clutter dynamic, the GA-RBF is also shown to enhance the detectability of small objects embedded in the sea. Using real-life radar data that are collected in the east coast of Canada by two different radar systems: a ground-based radar and a satellite equipped with synthetic aperture radar (SAR), we show that the GA-RBF network is a reliable detector for small surface targets in various sea conditions and is practical for real-life search and rescue, navigation, and surveillance applications.

New morphological detection algorithm based on the hit-miss transform

A new hit-miss transform (HMT) algorithm is proposed to de- tect distorted multiple objects in clutter. The HMT in morphology is used to locate a specific object in an input image. But the standard algorithm of the morphological HMT may cause problems in detecting various true- class objects. To provide efficient recognition of various true-class ob- jects in the input image, a new optical hit-miss morphological transform using the synthetic structuring element (SE) is proposed. The synthetic hit SE is composed of the intersection of all true-class hit SEs, and the synthetic miss SE is composed of the complement of the union of all true-class hit SEs. Simulation results show that the proposed algorithm can be used for the recognition of various true-class objects in an un- known input scene with only one HMT operation. © 2002 Society of Photo-

Polarized pulse waves in random discrete scatterers

In recent years there has been increasing interest in the use of polarization for imaging objects in a cluttered environment. Examples are optical imaging through clouds, optical detection of objects in a biological medium, and microwave detection of objects in clutter. We extend previous studies of continuous-wave scattering to pulse-polarization scattering in discrete scatterers. We solve the time-dependent vector radiative transfer equation for a plane-parallel medium by using Mie scattering and the discrete ordinates method. The time-dependent degree of polarization and cross-polarization discrimination are calculated and verify the advantages of circular over linear polarization in maintaining greater copolarized components rather than cross-polarized components.

Small target detection in clutter using recursive nonlinear prediction

Detecting small objects in clutter is usually carried out by using a predictor to suppress the background clutter. The idea is that a predictor which is trained using clutter data usually has small prediction error for the clutter process, but the prediction error will be relatively large if the signal fed into the predictor contains a target. While conventional approaches use a one-step-ahead predictor, we propose using a recursive predictor, which uses the predicted value to continue predicting the future points, to improve this predictive detection scheme. It is shown here that while the recursive prediction error of the clutter process is about the same as that of a one-step ahead predictor, the recursive predictor amplifies the prediction error of the target process. Therefore, the distance between the clutter and target processes is increased and the target detectability is enhanced. In addition, this recursive prediction approach has the same computational complexity as the one-step-ahead predictor since no extra training or modeling procedure is required. Real radar oceanic surveillance data are used to illustrate the effectiveness of the proposed detection method. Results show that the recursive prediction approach outperforms the one-step-ahead predictor in detecting small targets in the presence of strong clutter.

Contour Tracking in Clutter: A Subset Approach

A new method for tracking contours of moving objects in clutter is presented. For a given object, a model of its contours is learned from training data in the form of a subset of contour space. Greater complexity is added to the contour model by analyzing rigid and non-rigid transformations of contours separately. In the course of tracking, multiple contours may be observed due to the presence of extraneous edges in the form of clutters the learned model guides the algorithm in picking out the correct one. The algorithm, which is posed as a solution to a minimization problem, is made efficient by the use of several iterative schemes. Results applying the proposed algorithm to the tracking of a flexing finger and to a conversing individual's lips are presented.

Are cortical models really bound by the "binding problem"?

Are cortical models really bound by the "binding problem"?

Neural net design of macro Gabor wavelet filters for distortion-invariant object detection in clutter

We consider the detection of multiple classes of objects in clutter with 3-D object distortions and contrast differences present. We use a correlator because shift invariance is necessary to locate and recognize one object whose position is not known and to handle multiple objects in the same scene. The detection filter used is a linear combination of the real part of different Gabor filters, which we refer to as a macro Gabor filter (MGF). A new analysis of the parameters for the initial set of Gabor functions in the MGF is given, and a new neural network algorithm to refine these initial filter parameters and to determine the combination coefficients to produce the final MGF detection filter is detailed. Initial detection results are given. Use of this general neural network technique to design correlation filters for other applications seems very attractive.

Optical hit–miss morphological transform

We present an optical correlator implementation of the morphological hit-miss transform. This provides improved recognition of objects in clutter compared with standard optical pattern-recognition correlator techniques. The hit-miss transform is modified to use rank-order filtering since this gives better performance in noise and clutter. By varying the correlation plane threshold, we can perform dilations, rank-order filters, and erosions on the same multifunctional coherent optical correlator system. We quantify the thresholds required for generic object part recognition and provide simulated and optical laboratory data.