Browsing by Author "Chellappa, Rama"

Now showing 1 - 7 of 7
  • Thumbnail Image
    Item
    CFAR Detection of Targets in Fully Polarimetric SAR Images
    (1998-10-15) Wang, Ying; Chellappa, Rama; Zheng, Qinfen
    (Also cross-referenced as CAR-TR-696) Traditional constant false alarm rate (CFAR) detection algorithms produce a lot of false targets when applied to single-look, high-resolution, fully polarimetric synthetic aperture radar (SAR) images, due to the presence of speckle. We propose a two stag e CFAR detector followed by conditional dilation for detecting point and extended targets in polarimetric SAR images. In the first stage, possible targets are detected and false targets due to the speckle are removed by using global statistical parameters . In the second stage, the local statistical parameters are used to detect targets in regions adjacent to targets detected in the first stage. Conditional dilation is then performed to recover target pixels lost in second stage CFAR detection. The performance of a CFAR detector will be degraded if an incorrect statistical model is adopted and the data are correlated. A goodness-of-fit test is performed to decide the appropriate distribution and the effects of decorrelation of the data are cons idered. Good experimental results are obtained when our method is applied to single-look, highresolution, fully polarimetric SAR images acquired from MIT Lincoln Laboratory.
  • Thumbnail Image
    Item
    Estimation of Vehicle Dynamics from Monocular Noisy Images
    (1998-10-15) Yao, Yi-Sheng; Chellappa, Rama
    (Also cross-referenced as CAR-TR-692) This paper presents a new model-based egol lotion estimation algorithm for an autonomous vehicle navigating through rough terrain. Due to the uneven terrain, the vehicle undergoes bouncing, pitch and roll motion. To reliably accomplish other tasks such a s tracking and obstacle avoidance using visual inputs, it is essential to consider these disturbances. In this paper, two vehicle models available in the literature are used for egomotion estimation. The Half Vehicle Model (HVM) takes into account the bou ncing and pitch motion of the vehicle, and the Full Vehicle Model (FVM) also considers the roll motion. The dynamics of the vehicle are formulated using standard equations of motion. Assuming that depth information is known for some landmarks in the scene (e.g., obtained from a laser range finder), a feature-based approach is proposed to estimate vehicle motion parameters such as the vertical movement of the center of mass and the instantaneous angular velocity. An Iterated Extended Kalman Filter (IEKF) is used for recursive parameter estimation. Simulation results for both known and unknown terrain are presented.
  • Thumbnail Image
    Item
    Evaluation of Pattern Classifiers for Fingerprint and OCR Applications
    (1998-10-15) Blue, J.L.; Candela, G.T.; Grother, P.J.; Chellappa, Rama; Wilson, C.L.
    (Also cross-referenced as CAR-TR-691) In this paper we evaluate the classification accuracy of four statistical and three neural network classifiers for two image based pattern classification problems. These are fingerprint classification and optical character recognition (OCR) for isolated handprinted digits. The evaluation results reported here should be useful for designers of practical systems for these two important commercial applications. For the OCR problem, the Karhunen-Loeve (K-L) transform of the images is used to generate the inp ut feature set. Similarly for the fingerprint problem, the K-L transform of the ridge directions is used to generate the input feature set. The statistical classifiers used were Euclidean minimum distance, quadratic minimum distance, normal, and knearest neighbor. The neural network classifiers used were multilayer perceptron, radial basis function, and probabilistic. The OCR data consisted of 7,480 digit images for training and 23,140 digit images for testing. The fingerprint data consisted of 9,000 trai ning and 2,000 testing images. In addition to evaluation for accuracy, the multilayer perceptron and radial basis function networks were evaluated for size and generalization capability. For the evaluated datasets the best accuracy obtained for either pro blem was provided by the probabilistic neural network, where the minimum classification error was 2.5% for OCR and 7.2% for fingerprints.
  • Thumbnail Image
    Item
    An Improved Shape from Shading Algorithm
    (1998-10-15) Singh, Hemant; Chellappa, Rama
    (Also cross-referenced as CAR-TR-700) We propose an improved shape from shading (SFS) algorithm which is an extension of the recently published algorithm by Zheng and Chellappa [13]. A markedly more accurate estimate of the azimuth of the illumination source is presented. Depth reconstructio n has been improved upon by using a new set of boundary conditions and adapting a more sophisticated technique for hierarchical implementation of the SFS algorithm. Errors at the boundaries of images and in rotation of the reconstructed images have been c orrected. Typical results on synthetic and real images are presented.
  • Thumbnail Image
    Item
    Multiresolution Gauss Markov Random Field Models
    (1998-10-15) Krishnamachari, Santhana; Chellappa, Rama
    This paper presents multiresolution models for Gauss Markov random fields (GMRF) with applications to texture segmentation. Coarser resolution sample fields are obtained by either subsampling or local averaging the sample field at the fine resolution. Al though Markovianity is lost under such resolution transformation, coarse resolution non-Markov random fields can be effectively approximated by Markov fields. We present two techniques to estimate the GMRF parameters at coarser resolutions from the fine resolution parameters, one by minimizing the Kullback-Leibler distance and another based on local conditional distribution invariance. We show the validity of the estimators by comparing the power spectral densities of the Markov approximation and the exac t non-Markov measures. We also allude to the fact that different measures (different GMRF parameters) on the fine resolution can result in the same probability measure after subsampling and show the results for the first and second order cases. We apply this multiresolution model to texture segmentation. Different texture regions in an image are modeled by GMRFs and the associated parameters are assumed to be known. Parameters at lower resolutions are estimated from the fine resolution paramete rs. The coarsest resolution data is first segmented and the segmentation results are propagated upwards to the finer resolution. We use iterated conditional mode (ICM) minimization at all resolutions. A confidence measure is attached to the segmentation r esult at each pixel and passed on to the higher resolution. At each resolution, ICM is restricted only to pixels with low confidence measure. Our experiments with synthetic, Brodatz texture and real satellite images show that the multiresolution technique results in a better segmentation and requires lesser computation than the single resolution algorithm. (Also cross-referenced as UMIACS-TR-94-136)
  • Thumbnail Image
    Item
    Probe Based Recognition of Targets in Infrared Images
    (1998-10-15) Der, Sandor Z.; Chellappa, Rama
    (Also cross-referenced as CAR-TR-693) A probe based approach is used to recognize objects in a cluttered background using an infrared imager. A probe is a simple mathematical function which operates locally on image grey levels and produces an output that is more directly usable by an algori thm. A directional probe image is calculated by taking the difference in grey levels between pixels a set distance apart in a given direction, centered on the probe image pixel. These probe images contain the information necessary for use by an object rec ognition algorithm in a readily usable, and mathematically describable, form. A parametric statistical image background model which describes the probe images is introduced. The parameters of the probe image model can be readily estimated from the image. Knowledge of these parameters, together with target signatures obtained from Computer Aided Design (CAD) models, allows the likelihood ratio for a given object pose hypothesis versus the background null hypothesis to be written. The generalized likelihood ratio test is used to accept one of the object poses or to choose the null hypothesis. Results of the method applied to a large set of terrain model board images are presented.
  • Thumbnail Image
    Item
    Scalable Data Parallel Algorithms for Texture Synthesis and Compression using Gibbs Random Fields
    (1998-10-15) Bader, David A.; JaJa, Joseph; Chellappa, Rama
    This paper introduces scalable data parallel algorithms for image processing. Focusing on Gibbs and Markov Random Field model representation for textures, we present parallel algorithms for texture synthesis, compression, and maximum likelihood parameter estimation, currently implemented on Thinking Machines CM-2 and CM-5. Use of fine-grained, data parallel processing techniques yields real-time algorithms for texture synthesis and compression that are substantially faster than the previously known sequential implementations. Although current implementations are on Connection Machines, the methodology presented here enables machine independent scalable algorithms for a number of problems in image processing and analysis. (Also cross-referenced as UMIACS-TR-93-80.)