T. Kanungo, M. Y. Jaisimha, J. Palmer, and R. M. Haralick, "A methodology for quantitative performance evaluation of detection algorithms," IEEE Trans. Image Processing, vol. 4, pp. 1667--1674, 1995.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....[2] 3] Most of the objective evaluation methods assume knowledge of specific features such as known object boundaries in simple synthetic images. In such cases, the edge detection can be quantitatively measured, based on the known ideal detection considered to be the ground truth (GT) 4] 5] [6]. In real images, manual specification of the edges was applied to form a GT [7] 8] A method called local edge coherence measures local properties of continuation and thinness by examining local neighborhoods surrounding the detected edge points [9] and was later generalized to arbitrary ....

T. Kanungo, M.Y. Jaisimha, J. Palmer, and R.M. Haralick, "A Methodology for Quantitative Performance Evaluation of Detection Algorithms," IEEE Trans. Image Processing, vol. 4, no. 12, pp. 1667-1673, Dec. 1995.

....with q parameters, a total of 3 2 hypothesis tests can be performed. The authors propose different hypothesis tests when data comes from multivariate normal distribution. Kolmogorov Smirnov test is also described here that measures if two distributions are alike. Kanungo et al.[114] define the methodology for the quantitative performance evaluation of detection algorithms. A typical detection task relates to the identification of a target in an image. The task is to compute a number called evidence strength that measures the evidence whether the target is present. The ....

T. Kanungo, M.Y. Jaisimha, J. Palmer and R.M. Haralick, A methodology for quantitative performance evaluation of detection algorithms, IEEE Transactions on Image Processing, vol. 4, no. 12, pp. 1667-1674, 1995.

....have presented methods for empirical performance evaluation. For example, Hoover et al. 13] proposed an experimental framework for quantitative comparison of range image segmentation algorithms and demonstrated the methodology by evaluating four range segmentation algorithms. Kanungo et al. [17] described a four step methodology for the evaluation of two detection algorithms. Phillips and Chhabra [32] presented a methodology for empirically evaluating graphics recognition systems. These methodologies have not addressed the issues of automatic training of algorithms with free parameters ....

.... default parameter values are selected and no training method is explicitly specified [19, 2, 28, 15, 30, 7] Similarly, in performance evaluation literature where the algorithm parameters can be set by evaluators, a set of parameter values are usually selected manually in the training procedure [13, 17, 32]. A common aspect of these parameter value selection methods and training methods is that a set of optimal parameter values are manually selected based on some assumption regarding the training dataset. To objectively optimize a segmentation algorithm on a given training dataset, a set of ....

[Article contains additional citation context not shown here]

T. Kanungo, M. Y. Jaisimha, J. Palmer, and R. M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE Transactions on Image Processing, 4:1667--1674, 1995.

....analysis of their performance, but rely or resort instead to a qualitative demonstration, merely plotting their resulting segmentation. Naturally, this is unsatisfactory since it is difficult to assess the relative merits of the various algorithms, and a more quantitative approach is necessary [11]. Fischler and Wolf [6] rated curve segmentation results using human observers. However, a more convenient and repeatable approach would be preferable. Several recent papers on dominant point detection have quantified their performance based on: the percentage of missed points versus the ....

T. Kanungo, M.Y. Jaisimha, J. Palmer, and R.M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE IP, 4:1667--1673, 1995.

....Performance of Applications: The fourth class consists of measures that directly assess the performance of applications. Examples include, the computation of projective invariants [24] object recognition [47] structure from motion [112] industrial inspection [114] and boundary extraction [56] [90] Local Measures of Coherence: The fifth and final class consists of measures that are based upon desirable local properties of the output feature map, for example, continuation and thinness [57] 95] 121] Such properties are particularly important for applications that first aggregate ....

T. Kanungo, M.Y. Jaisimha, J. Palmer, and R.M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE Transactions on Image Processing, 4(12):1667--1673, December 1995. 160

....imaging community is generally unconvinced by the type of anecdotal single image comparisons often found in image processing papers. There is increasing emphasis on formal statistical evaluations of different image reconstruction methods [49 51] which are also being applied to image processing [52]. J. To do (for tech report) bias variance for ML SAGE post smoothed (ala sieves) vs PML SAGE. 6 ....

T Kanungo, M Y Jaisimha, J Palmer, and R M Haralick. A methodology for quantitative performanceevaluation of detection algorithms. IEEE Tr. Im. Proc., 4(12):1667--1674, December 1995.

....analysis of their performance, but rely or resort instead to a qualitative demonstration, merely plotting their resulting segmentation. Naturally, this is unsatisfactory since it is difficult to assess the relative merits of the various algorithms, and a more quantitative approach is necessary [10, 13]. Fischler and Wolf [7] rated curve segmentation results using human observers. However, a more convenient and repeatable approach would be preferable. Several recent papers on dominant point detection have quantified their performance based on: the percentage of missed points versus the ....

T. Kanungo, M.Y. Jaisimha, J. Palmer, and R.M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE Trans. Image Processing, 4:1667--1673, 1995.

....specifications but rather whether perceived edges are detected. Subjective evaluations are vague and cannot be used to measure the performance of detectors but only to establish their failure. The goal of objective evaluation is to measure the performance of an edge detector. Several authors [1, 88, 49, 115, 46] have proposed performance measures to evaluate the output of edge detectors. Abdou and Pratt [1, 88] have proposed a measure, called the figure of merit, which is a combination of three factors: non detection of true edges, detection of false edges, and edge delocalization error. Using this ....

....same interest as the smoothing and differentiation techniques. We suggest that evaluation methods should take into account the subsequent use of edges, the specification of the detector and the characteristics of the real image. Recent results obtained by Heath et al. 39] and by Kanungo et al. [46] are a promising step in this direction. 5 Survey of Edge Detectors Since the appearance of image processing, the number of edge detectors has increased continuously. It is difficult to make an inventory of the available algorithms. Most existing edge detectors are autonomous and include the ....

T. Kanungo, M.Y. Jaisimba, J. Palmer, and R.M. Haralick. A Methodology for Quantitative Performance Evaluation of Detection Algorithms. ieee Transactions on Image Processing, 12(4):1667--1674, 1995.

No context found.

T. Kanungo, M. Y. Jaisimha, J. Palmer, and R. M. Haralick, "A methodology for quantitative performance evaluation of detection algorithms," IEEE Trans. Image Processing, vol. 4, pp. 1667--1674, 1995.

....have presented methods for empirical performance evaluation. For example, Hoover et al. 13] proposed an experimental framework for quantitative comparison of range image segmentation algorithms and demonstrated the methodology by evaluating four range segmentation algorithms. Kanungo et al. [17] described a four step methodology for the evaluation of two detection algorithms. Phillips and Chhabra [32] presented a methodology for empirically evaluating graphics recognition systems. These methodologies have not addressed the issues of automatic training of algorithms with free parameters ....

.... default parameter values are selected and no training method is explicitly specified [19, 2, 28, 15, 30, 7] Similarly, in performance evaluation literature where the algorithm parameters can be set by evaluators, a set of parameter values are usually selected manually in the training procedure [13, 17, 32]. A common aspect of these parameter value selection methods and training methods is that a set of optimal parameter values are manually selected based on some assumption regarding the training dataset. To objectively optimize a segmentation algorithm on a given training dataset, a set of ....

[Article contains additional citation context not shown here]

T. Kanungo, M. Y. Jaisimha, J. Palmer, and R. M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE Transactions on Image Processing, 4:1667--1674, 1995.

No context found.

T. Kanungo, M. Y. Jaisimha, J. Palmer, and R. M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE Trans. on Image Processing, 4:1667--1674, 1995.

....gain matrix (see text) Radiology community (see [3] and literature cited therein) and automatic target recognition (ATR) community make use of ROC methodology. For a discussion of quantitative performance evaluation for line detection algorithms where optimal operating points can be used see [4, 5]. 2. PRACTICAL ISSUES Although the theory described in the previous section has existed in the statistical signal detection literature for decades [1] and has been used by the radiology and ATR community quite extensively, there is a practical problem that still remains: Since the ROC data is ....

T. Kanungo, M.Y. Jaisimha, J. Palmer, and R.M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE Trans. on Image Processing (to appear), December 1994.

....and results on sample data sets are given. 1. INTRODUCTION In this paper we consider two problems where monotonic curve fitting with endpoint constraints is necessary. First problem comes about while fitting a function to data points that represent the operating characteristics of a system (see [1, 2, 3]) Since the receiver operating characteristic (ROC) curve is plot of probability of misdetection versus the probability of false alarm, the data points are monotonically decreasing, and the fitted function also needs to be monotonically decreasing. The second problem is that of fitting smooth ....

T. Kanungo, M. Y. Jaisimha, J. Palmer, and R. M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE Trans. on Image Processing (to appear), December 1994.

....from one image. Thus, more depth first (end to end) research is currently required as opposed to breadth first (complexity of objects, line drawings, etc) DK93] Second, criterion for evaluation and complete experimental protocols using the specified criterion should be used and reported [Har89, KJPH94, KJPH93] This will enable the scientific community to replicate results reported in the literature. The performance evaluation should be based on a reasonably large set of simulated and real engineering drawings, which the system is supposed to process and understand. Many experimental systems ....

T. Kanungo, M. Y. Jaisimha, J. Palmer, and R. M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE Trans. on Image Processing (to appear), 1994.

No context found.

T. Kanungo, M.Y. Jaisimha, J. Palmer, and R.M. Haralick. A methodology for quantitative performance evaluation of detection algorithms. IEEE Transactions on Image Processing (to appear), 1993. 107

No context found.

T. Kanungo, M.Y. Jaisimha, J. Palmer, and R.M. Haralick, "A Methodology for Quantitative Performance Evaluation of Detection Algorithms," IEEE Trans. Image Processing, vol. 4, no. 12, pp. 1667-1673, Dec. 1995.

No context found.

T Kanungo, M Y Jaisimha, J Palmer, and R M Haralick, "A methodology for quantitative performance evaluation of detection algorithms," IEEE Trans. Image Process. 4 1667--1674 (1995).

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC