## 10 Pros and Cons Against Performance Characterization of Vision Algorithms (1996)

Citations: | 33 - 4 self |

### BibTeX

@INPROCEEDINGS{Förstner9610pros,

author = {Wolfgang Förstner},

title = {10 Pros and Cons Against Performance Characterization of Vision Algorithms},

booktitle = {},

year = {1996},

pages = {13--29}

}

### Years of Citing Articles

### OpenURL

### Abstract

The paper discusses objections against performance characterization of vision algorithms and explains their motivation. Short and long-term arguments are given which overcome these objections. The methodology for performance characterization is sketched to demonstrate the feasibility of empirical testing staffof vision algorithms. 0 Motivation For at least 10 years Computer Vision has been confronted with papers and discussions on the scientific value of its results and the difficulties in transferring the results to practical systems. A change of awareness seems to have happened: At the Computer Vision Workshop 1985 two controversial papers, with different views, agreed on the lack of theoretical research ([9], [13]), which should go along with the development of vision procedures: experimental proofs are not enough. The dialogue on 'Ignorance, Myopia, and Naivit'e in Computer Vision Systems' initiated by R. Jain and T. Binford ([10]) and the responses documented the necessity of ev...

### Citations

388 |
Estimating uncertain spatial relationships in robotics
- Smith, Self, et al.
- 1990
(Show Context)
Citation Context ...e to the chosen coordinate system: In spite of the same coordinates being given, their variances are not comparable as they are given in different reference systems. The solution goes back to [2] but =-=[14]-=- have independently identified and solved the problem in the context of robotics (cf. also [15]). Assume a robot is able to measure the length of its path between two positions, while walking on a str... |

131 | A Stochastic Map for Uncertain Spatial Relationships
- Smith, Self, et al.
- 1987
(Show Context)
Citation Context ...s are not comparable as they are given in different reference systems. The solution goes back to [2] but [14] have independently identified and solved the problem in the context of robotics (cf. also =-=[15]-=-). Assume a robot is able to measure the length of its path between two positions, while walking on a straight line. Starting at position P1(x1) it moves to point P2(x2) and then to position P3 (x3 ).... |

100 |
A framework for low-level feature extraction
- Förstner
- 1994
(Show Context)
Citation Context ...fi(ffi; ff) (12) An example for an empirically derived line coverage is given in Fig. 1. It is taken from [8], where a complete analysis of the quality of a polymorphic feature extraction scheme (cf. =-=[7]-=-) is given. Figure 1: The expected coverage c(L) of an extracted edge in dependency on the signal to noise ratio SNR = contrast=oe n . For SNR ? 2.5 one can expect the line to be extracted without gap... |

29 |
Reliability analysis of parameter estimation in linear models with applications to mensuration problems in computer vision
- Förstner
- 1987
(Show Context)
Citation Context ...e test, and the true edge leads to a gradient magnitude being a factor ffi (non-centrality parameter of the non-central Normal-distribution) larger than oe gu , the power of the test is given by (cf. =-=[4]-=-): fi(ffi; ff) = P (jT j ? kjpixel is edge) = 1 \Gamma \Phi(k(ff) \Gamma ffi) + \Phi(k(ff) + ffi) with the normalized Gaussian distribution \Phi(x). The standard deviation oe gu of the gradient magnit... |

21 |
Computer Vision Theory: The Lack Thereof
- Haralick
- 1985
(Show Context)
Citation Context ...ts to practical systems. A change of awareness seems to have happened: At the Computer Vision Workshop 1985 two controversial papers, with different views, agreed on the lack of theoretical research (=-=[9]-=-, [13]), which should go along with the development of vision procedures: experimental proofs are not enough. The dialogue on 'Ignorance, Myopia, and Naivit'e in Computer Vision Systems' initiated by ... |

9 |
Diagnostics and Performance Evaluation in Computer Vision
- Forstner
- 1994
(Show Context)
Citation Context ...from (13) is independent on the chosen reference variance, thus would yield the same value if the new weights would have been chosen. Experiences with longer chains of image analysis steps (cf. e. g. =-=[6]-=-) confirm the possibility to link suboptimal partial results without losing the ability to evaluate the final result, in spite of the submodules containing severe nonlinearities. 4 The used models are... |

8 | On the Noise and Scale Behaviour of Relational Descriptions
- Fuchs, Forstner, et al.
- 1994
(Show Context)
Citation Context ... length of the line in pixels and is given by c(L) = #edge pixels found length of edge[pel] = fi(ffi; ff) (12) An example for an empirically derived line coverage is given in Fig. 1. It is taken from =-=[8]-=-, where a complete analysis of the quality of a polymorphic feature extraction scheme (cf. [7]) is given. Figure 1: The expected coverage c(L) of an extracted edge in dependency on the signal to noise... |

6 |
I've seen your demo; so what
- Price
- 1985
(Show Context)
Citation Context ... practical systems. A change of awareness seems to have happened: At the Computer Vision Workshop 1985 two controversial papers, with different views, agreed on the lack of theoretical research ([9], =-=[13]-=-), which should go along with the development of vision procedures: experimental proofs are not enough. The dialogue on 'Ignorance, Myopia, and Naivit'e in Computer Vision Systems' initiated by R. Jai... |

5 |
S-Transformations and Criterion Matrices, volume Band 5 der Reihe 1, Netherlands Geodetic Commission
- Baarda
- 1973
(Show Context)
Citation Context ...reference to the chosen coordinate system: In spite of the same coordinates being given, their variances are not comparable as they are given in different reference systems. The solution goes back to =-=[2]-=- but [14] have independently identified and solved the problem in the context of robotics (cf. also [15]). Assume a robot is able to measure the length of its path between two positions, while walking... |

4 | Automated learning of rules using genetic operators
- Schnier, Bloemer
- 1993
(Show Context)
Citation Context ...ection of an admissible set out of d np possible parameter values. A clear consequence is to reduce tuning parameters to a minimum. This would also help automated systems for parameter selection (cf. =-=[12]-=-) As goal/task driven control of algorithms still is necessary, a small set of tuning parameters may be left. But only tuning parameters with a well-defined meaning for control should be allowed. Exam... |

4 | Parameterfree Information-Preserving Surface Restoration
- Weidner
- 1994
(Show Context)
Citation Context ...variance component estimation (cf. [3]), no free parameter is necessary in regularization. Of course, making the variances dependent on the position allows adaptation to any type of irregularity (cf. =-=[16]). Moreove-=-r, the regularization term suggests the profile to follow an autoregressive scheme of order 2, namely f i+1 = 2f i \Gamma f i\Gamma1 + " i with oe " i = oes, allowing to generalize the regul... |

3 |
Noise estimation for color edge extraction
- Brugelmann, Forstner
- 1992
(Show Context)
Citation Context ... the precision oe g of the observed values and the roughness oesof the reconstructed function to determine . As both variances can be estimated from real data using variance component estimation (cf. =-=[3]-=-), no free parameter is necessary in regularization. Of course, making the variances dependent on the position allows adaptation to any type of irregularity (cf. [16]). Moreover, the regularization te... |

2 |
Dialogue--Ignorance, myopia, and naivit`e in computer vision systems. CVGIP: Image Understanding 53
- Jain, Binford
- 1991
(Show Context)
Citation Context ... along with the development of vision procedures: experimental proofs are not enough. The dialogue on 'Ignorance, Myopia, and Naivit'e in Computer Vision Systems' initiated by R. Jain and T. Binford (=-=[10]-=-) and the responses documented the necessity of evaluating theoretical findings, vision procedures algorithms etc. by using empirical data in order to increase the number of real world applications of... |

2 | S-Transformations and Criterion Matrices, Band 5 der Reihe 1. Netherlands Geodetic Commission - Baarda - 1973 |

1 |
On the theoretical accuracy of planimetric block triangulation
- Ackermann
- 1966
(Show Context)
Citation Context ...med within two years. The survey agencies strongly supported the university as there was an interest in the results. The reason for the interest was motivated by the theoretical research going ahead (=-=[1]-=-) or parallel indicating the power of the proposed methods. 9 Simulations are not reality 9.1 Pro Simulations cannot replace reality, they always are too ideal. As generating artificial images uses th... |

1 |
Uncertain spatial relationships and their use for object location in digital images
- Forstner
- 1992
(Show Context)
Citation Context ...eviation of the orientation. The precision of an edge can easily be derived in case we treat edge location as template matching. We obtain for the variance oe 2 u of the position across the edge (cf. =-=[5]-=-) oe 2 u = oe 2 n P r;c f 2 u (r; c) (5) where oe n is the standard deviation of the noise, fu is the derivative of the template edge across the edge, i. e. in u-direction, and the sum is to be taken ... |

1 |
Die katasterphotogrammetrie im praktischen einsatz
- Kraus
- 1973
(Show Context)
Citation Context ...other hand. In the early 70's, a number of Photogrammetric software packages for the simultaneous orientation of large sets of aerial images were developed. In a tutorial on these developments Kraus (=-=[11]-=-) reported on 30 projects with between 350 and 7140 unknown points in object space, total 52250. They were manually checked in the field. The total cost per point was estimated to be in the range betw... |

1 | I've SeenYour Demo� So What? In - Price - 1985 |