Image analysis – Applications – Target tracking or detecting
Reexamination Certificate
1999-06-17
2002-05-21
Johns, Andrew W. (Department: 2721)
Image analysis
Applications
Target tracking or detecting
C382S240000
Reexamination Certificate
active
06393137
ABSTRACT:
BACKGROUND OF THE INVENTION
1. Field of the Invention
The present invention relates to methods and systems for feature extraction and object classification. More specifically, the present invention relates to multi-resolution signal processing methods and systems for extraction and classification of embedded features.
2. Description of the Related Art
Standard object classification methodology begins with a “physical world,” composed of the object classes that the user desires to discriminate. For example, the standard missile-3 (SM-3) Advanced Light-Weight Exo-Atmospheric Projectile (LEAP) infrared (IR) seeker takes measurements (images) of these objects and the raw imagery is then pre-processed using a collection of algorithms. At this point, the data space is highly redundant. The data is then mapped by some unitary transformation, e.g., Principal Components, Spectral Decomposition, Time-Frequency Decomposition, etc. This so-called feature space will in general have a greatly reduced dimensionality. The feature set is then used as input for a classifier. This process is generally described in U.S. Pat. No. 5,093,869, which is incorporated herein by reference.
The primary objective of the object classifier is to discriminate the reentry vehicle (RV) from the other three object classes. A secondary goal of the object classifier is to reduce misclassifications among the remaining objects.
It will be appreciated that since all of the observed objects in the TBM mission will be sub-pixel, only the temporal information is available for the discrimination task. Therefore, an accumulating time period is needed to collect the temporal data before initiating the discrimination task. A well-designed object classifier would use a shorter temporal data length and, thus, a reduced accumulating time period, but still provide good classification performance.
The most commonly employed approaches to the feature extraction problem are based on either Fourier analysis or eigenvalue decomposition methods. It will be appreciated that the current Feature Extraction algorithm of choice for SM-3 LEAP designs is based on Power Spectral Densities (PSDs). Of critical importance, SM-3 LEAP systems must be able to maintain their current classifier design, since the classifier is very sophisticated and has shown excellent performance against simulated data sets.
As mentioned above, one possible algorithm for TBM target discrimination is a Fourier technique based on the construction of PSD templates. Moreover, the main goal is to separate the RV from the other three object classes, i.e., an associated object, a booster, and burnt solid fuel. At a very high level, the general procedure is to pre-compute and store—for each object class—the averaged PSD templates as well as the mean and covariance matrices for several thousand Monte-Carlo simulation runs. From an analysis of these templates, it is possible to identify features that can be used to construct the feature vector. For each object observed, a PSD is generated, compared to the pre-computed PSD template for the RV and a feature vector is formed. This feature vector is then used to calculate the discriminant function, g
i
, for each object class. The largest of these discriminant functions then corresponds to the object class.
Hence, there is a need in the art for a multi-resolution feature extraction method and corresponding system providing a greatly expanded feature space. It would be desirable if the expanded feature space produced by the multi-resolution feature extraction method and corresponding system could be efficiently and adaptively calculated. What is also needed is a multi-resolution feature extraction method and system which permits generation of a relatively higher quality of extracted features, which leads directly to improved classifier performance. It would be beneficial if the classifier could be simplified by virtue of the higher quality of the extracted features.
SUMMARY OF THE INVENTION
The need in the art is addressed the multi-resolution feature extraction method and corresponding apparatus of the present invention. In the illustrative embodiment, the feature extractor includes circuitry for receiving and transforming a time variant data signal into a multi-resolution data signal. The multi-resolution data signal is compared to each of a plurality of object templates. The system then generates a feature vector based on a correlation of the multi-resolution data signal to one of the object templates.
The multi-resolution feature extraction method according to the present invention, employs object templates formed by transforming time variant image data for each of a plurality of objects into a respective multi-resolution template and averaging all templates for each respective object. The inventive method includes steps for transforming an incoming time variant data signal into a multi-resolution data signal, comparing the multi-resolution data signal to each of the object templates, and generating a feature vector when the multi-resolution data signal correlates to one of the object templates.
In a more specific implementation, the inventive method further includes the steps of calculating a confusion matrix (CM), classifying the feature vectors as one of the objects to thereby produce classified objects responsive to the CM, and selecting a target from the classified objects. Preferably, the method also includes a step for, when there are more than one of the objects in the incoming time variant data signal, computing a probability-of-error vector (PV), so that the selecting step can select a target from the classified objects responsive to the PV.
Additionally, a multi-resolution feature extractor according to the present invention employs object templates formed by transforming time variant image data for each of a plurality of objects into a respective multi-resolution template and averaging all templates for each respective object to thereby generate object templates. Preferably, the multi-resolution feature extractor includes a memory which stores the object templates, and a feature extractor which transforms an incoming time variant data signal into a multi-resolution data signal, compares the multi-resolution data signal to each of the object templates, and generates a feature vector when the multi-resolution data signal correlates to one of the object templates.
Moreover, a multi-resolution seeker system constructed in accordance with the present teachings includes a memory and a feature extractor operatively connected thereto. The system converts time variant image data for each of a plurality of objects into a respective multi-resolution template, averages all templates for each respective object to thereby generate object templates, transforms an incoming time variant data signal into a multi-resolution data signal, compares the multi-resolution data signal to each of the object templates, and generates a feature vector each time the multi-resolution data signal correlates to one of the object templates. A processor calculates a confusion matrix (CM). A classifier operatively connected to the memory classifies the feature vectors as one of the objects responsive to the CM to thereby produce classified objects. A selector operatively coupled to the memory selects a target from the classified objects.
According to one aspect of the inventive system, the processor, when there are more than one of the objects in the incoming time variant data signal, computes a probability-of-error vector (PV), to permit the selector to select the target from the classified objects responsive to the PV.
REFERENCES:
patent: 5063603 (1991-11-01), Burt
patent: 6154567 (2000-11-01), McGarry
patent: 6292592 (2001-09-01), Braunreiter et al.
Chen Hai-Wen
Riddle Jack G.
Schmitt Harry A.
Azarian Seyed
Benman William J.
Johns Andrew W.
Lenzen, Jr. Glenn H.
Raufer Colin M.
LandOfFree
Multi-resolution object classification method employing... does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Multi-resolution object classification method employing..., we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Multi-resolution object classification method employing... will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-2912451