Browsing by Author "Ghosal, Sugata"

Now showing 1 - 13 of 13

Adaptable Similarity Search using Non-Relevant Information
(Elsevier, 2002-08) Ghosal, Sugata
This chapter presents a novel technique for improving the accuracy of adaptable similarity based retrieval by incorporating negative relevance judgment, and demonstrates excellent performance and robustness of the proposed scheme with a large number of experiments. Many modern database applications require content-based similarity search capability in numeric attribute space. Therefore, online techniques for adaptively refining the similarity metric based on relevance feedback from the user are necessary. Existing methods use retrieved items marked relevant by the user to refine the similarity metric, without taking into account the information about non-relevant (or unsatisfactory) items. Consequently, items in database close to non-relevant ones continue to be retrieved in further iterations. A decision surface is determined to split the attribute space into relevant and non-relevant regions. The decision surface is composed of hyperplanes, each of which is normal to the minimum distance vector from a non-relevant point to the convex hull of the relevant points.
Automatic substructuring for domain decomposition using neural networks
(IEEE, 2002-08) Ghosal, Sugata
Application of neural networks for guiding solutions of large numerical problems is an emerging area of research. Automatic generation of subdomains from large 3D finite element meshes is a key preprocessing step in domain decomposition techniques and extremely important for proper load balancing, reducing communication bandwidth and latency, and efficient processor coordination and synchronization in a parallel computing environment. It is desired that the subdomains are approximately of same size, and the total number of interface nodes between adjacent subdomains is minimal. We propose two neural network algorithms employing the philosophy of competitive learning and Hopfield network, that can automatically generate substructures from large 3D meshes with reasonable speed. Both these techniques are implemented in such as a way that they have almost linear complexity w.r.t. the problem size for serial execution. Experimental results show more than 25% improvement over an existing greedy algorithm
Detection of composite edges
(IEEE, 1994-01) Ghosal, Sugata
The paper presents a new parametric model-based approach to high-precision composite edge detection using orthogonal Zernike moment-based operators. It deals with two types of composite edges: (a) generalized step and (b) pulse/staircase edges. A 2-D generalized step edge is modeled in terms of five parameters: two gradients on two sides of the edge, the distance from the center of the candidate pixel, the orientation of the edge and the step size at the location of the edge. A 2-D pulse/staircase edge is modeled in terms of two steps located at two positions within the mask, and the edge orientation. A pulse edge is formed if the steps are of opposite polarities whereas a staircase edge results from two steps having the same polarity. Two complex and two real Zernike moment-based masks are designed to determine parameters of both the 2-D edge models. For a given edge model, estimated parameter values at a point are used to detect the presence or absence of that type of edge. Extensive noise analysis is performed to demonstrate the robustness of the proposed operators. Experimental results with intensity and range images are included to demonstrate the efficacy of the proposed edge detection technique as well as to compare its performance with the geometric moment-based step edge detection technique and Canny's (1986) edge detector
A fast scalable algorithm for discontinuous optical flow estimation
(IEEE, 1996-02) Ghosal, Sugata
Multiple moving objects, partially occluded objects, or even a single object moving against the background gives rise to discontinuities in the optical flow field in corresponding image sequences. While uniform global regularization based moderately fast techniques cannot provide accurate estimates of the discontinuous flow field, statistical optimization based accurate techniques suffer from excessive solution time. A 'weighted anisotropic' smoothness based numerically robust algorithm is proposed that can generate discontinuous optical flow field with high speed and linear computational complexity. Weighted sum of the first-order spatial derivatives of the flow field is used for regularization. Less regularization is performed where strong gradient information is available. The flow field at any point is interpolated more from those at neighboring points along the weaker intensity gradient component. Such intensity gradient weighted regularization leads to Euler-Lagrange equations with strong anisotropies coupled with discontinuities in their coefficients. A robust multilevel iterative technique, that recursively generates coarse-level problems based on intensity gradient weighted smoothing weights, is employed to estimate discontinuous optical flow field. Experimental results are presented to demonstrate the efficacy of the proposed technique.
An image retrieval system with automatic query modification
(IEEE, 2002-06) Ghosal, Sugata
Most interactive "query-by-example" based image retrieval systems utilize relevance feedback from the user for bridging the gap between the user's implied concept and the low-level image representation in the database. However, traditional relevance feedback usage in the context of content-based image retrieval (CBIR) may not be very efficient due to a significant overhead in database search and image download time in client-server environments. In this paper, we propose a CBIR system that efficiently addresses the inherent subjectivity in user perception during a retrieval session by employing a novel idea of intra-query modification and learning. The proposed system generates an object-level view of the query image using a new color segmentation technique. Color, shape and spatial features of individual segments are used for image representation and retrieval. The proposed system automatically generates a set of modifications by manipulating the features of the query segment(s). An initial estimate of user perception is learned from the user feedback provided on the set of modified images. This largely improves the precision in the first database search itself and alleviates the overheads of database search and image download. Precision-to-recall ratio is improved in further iterations through a new relevance feedback technique that utilizes both positive as well as negative examples. Extensive experiments have been conducted to demonstrate the feasibility and advantages of the proposed system.
Leveraging non-relevant images to enhance image retrieval performance
(ACM Digital Library, 2002-12) Ghosal, Sugata
Inherent subjectivity in user's perception of an image has motivated the use of relevance feedback (RF) in the image desigined output's retrieval process. RF techniques interactively determine the user's query concept, given the user's relevance judgments on a set of images. In this paper we propose a robust technique that utilizes non-relevant images to efficiently discover the relevant search region. A similarity metric, estimated using the relevant images is then used to rank and retrieve database images in the relevant region. The partitioning of the feature space is achieved by using a piecewise linear decision surface that separates the relevant and non-relevant images. Each of the hyperplanes constituting the decision surface is normal to the minimum distance vector from a non-relevant point to the convex hull of relevant points. Experimental results demonstrate significant improvement in retrieval performance for the small feedback size scenario over two well established RF algorithms.
A moment-based unified approach to image feature detection
(IEEE, 1997-06) Ghosal, Sugata
In this paper, a novel model-based approach is proposed for generating a set of image feature maps (or primal sketches). For each type of feature, a piecewise smooth parametric model is developed to characterize the local intensity function in an image. Projections of the intensity profile onto a set of orthogonal Zernike-moment-generating polynomials are used to estimate model-parameters and, in turn, generate the desired feature map. A small set of moment-based detectors is identified that can extract various kinds of primal sketches from intensity as well as range images. One main advantage of using parametric model-based techniques is that it is possible to extract complete information (i.e., model parameters) about the underlying image feature, which is desirable in many high-level vision tasks. Experimental results are included to demonstrate the effectiveness of proposed feature detectors.
Orthogonal moment operators for subpixel edge detection
(Elsevier, 1993-02) Ghosal, Sugata
A new approach to detect step edges with subpixel accuracy is presented. The proposed approach is based on a set of orthogonal complex moments of the image known as Zernike moments. An ideal two-dimensional (2D) step edge is modeled in terms of four parameters: the background gray level, the step size, the distance of the edge from the center of the mask, and the orientation of the edge. Discrete Zernike moments are used to obtain a total of three complex masks to compute all the edge parameters for subpixel detection. For pixel-level edge detection only two masks (one real and one complex) are required. The theoretical analysis of the influence of noise on the location and the orientation of an edge is presented. This analysis reveals that the accuracy of the proposed approach is virtually unaffected by the additive noise. The technique is effective in detecting both the pixel-level and subpixel-level edges. Experimental results are presented to demonstrate the efficacy of the proposed technique.
QoSMig: adaptive rate-controlled migration of bulk data in storage systems
(IEEE, 2005-04) Ghosal, Sugata
Logical reorganization of data and requirements of differentiated QoS in information systems necessitate bulk data migration by the underlying storage layer. Such data migration needs to ensure that regular client I/Os are not impacted significantly while migration is in progress. We formalize the data migration problem in a unified admission control framework that captures both the performance requirements of client I/Os and the constraints associated with migration. We propose an adaptive rate-control based data migration methodology, QoSMig, that achieves the optimal client performance in a differentiated QoS setting, while ensuring that the specified migration constraints are met QoSMig uses both long term averages and short term forecasts of client traffic to compute a migration schedule. We present an architecture based on Service Level Enforcement Discipline for Storage (SLEDS) that supports QoSMig. Our trace-driven experimental study demonstrates that QoSMig provides significantly better I/O performance as compared to existing migration methodologies
Range surface characterization and segmentation using neural networks
(Elsevier, 1995) Ghosal, Sugata
This paper presents an integrated neural net-based approach to the segmentation of range images into distinct surfaces, which is an essential step in range image analysis and interpretation. A two-stage connectionist neural net model is proposed which extracts local surface features at each image point and groups pixels via local interactions among different features. The first stage computes surface parameters, e.g., surface normals, curvature and discontinuities (crease and jump) by optimally projecting the local range profile onto a set of non-orthogonal basis functions. In the second stage, adjacent pixels compete with each other based on the surface features associated with them to group themselves into different surface patches. Daugman's projection neural net (DPNN) and Kohonen's self-organizing neural net (KSNN) are used for the feature extraction and region-growing, respectively. Empirical performance analysis shows that the feature extraction using neural net is quite robust with respect to the additive noise. Experimental results are included to demonstrate the performance of the proposed technique.
Segmentation of range images: an orthogonal moment-based integrated approach
(IEEE, 1993) Ghosal, Sugata
A new approach to range image segmentation is presented. The proposed approach involves two phases in which the region and edge information detected using a set of orthogonal Zernike moment-based operators are combined to provide robust segmentation of range images. In the first phase, each range image point is characterized by the surface normal vector and the depth value at that point. A surface feature-based clustering of range image points yields its initial region-based segmentation. This initial segmentation phase often produces oversegmented images. In the second phase of the proposed technique, the oversegmented image is resegmented by appropriately merging adjacent regions using the edge information to produce final segmentation. One attractive characteristic of the proposed technique is that the same set of three moment-based operators is used to extract both surface and edge features. Thus only three convolution operations are needed at an image point to compute all the desired surface and edge features associated with that point. The performances of the proposed Zernike moment-based operators in surface and edge feature detection are theoretically analyzed
Target detection in foveal ATR systems
(IEEE, 1996) Ghosal, Sugata
Automatic target recognition (ATR) applications require simultaneously a wide field of view (FOV) for better detection and situation awareness, high resolution for target recognition and threat assessment, and high frame rate for detecting brief events and disambiguating frame-to-frame correlation. Uniformly sampling the entire FOV at recognition resolution is simply wasteful in ATR scenarios with localized regions of interest (ROIs). Foveal data acquisition with space-variant sampling and context-sensitive sensor articulation is highly optimized for active ATR applications. We propose a multiscale local Zernike filter-based front end target detection technique for a commercially feasible foveal sensor topology with piecewise constant resolution profile. Anisotropic heat diffusion is employed for preprocessing of the foveal data. Expansion template matching is used to derive a detection filter that optimizes the discriminant signal-to-noise ratio (SNR). Results are presented with simulated foveal imagery, derived from real uniform acuity FLIR data.
A utility-based unified disk scheduling framework for shared mixed-media services
(ACM Digital Library, 2008-02) Ghosal, Sugata
We present a new disk scheduling framework to address the needs of a shared multimedia service that provides differentiated multilevel quality-of-service for mixed-media workloads. In such a shared service, requests from different users have different associated performance objectives and utilities, in accordance with the negotiated service-level agreements (SLAs). Service providers typically provision resources only for average workload intensity, so it becomes important to handle workload surges in a way that maximizes the utility of the served requests. We capture the performance objectives and utilities associated with these multiclass diverse workloads in a unified framework and formulate the disk scheduling problem as a reward maximization problem. We map the reward maximization problem to a minimization problem on graphs and, by novel use of graph-theoretic techniques, design a scheduling algorithm that is computationally efficient and optimal in the class of seek-optimizing algorithms. Comprehensive experimental studies demonstrate that the proposed algorithm outperforms other disk schedulers under all loads, with the performance improvement approaching 100% under certain high load conditions. In contrast to existing schedulers, the proposed scheduler is extensible to new performance objectives (workload type) and utilities by simply altering the reward functions associated with the requests.