A Non-Linear Quantitative Evaluation

Approach for Disparity Estimation

Iván Cabezas and María Trujillo

VISAPP 2011

Conventional quantitative evaluation approaches for disparity estimation rely on the use of a single value as an indicator of comparative

performance. A non-linear quantitative evaluation approach for disparity estimation is proposed and supported by Pareto Dominance and Pareto

Optimal Set concepts. The proposed approach allows different evaluation scenarios and offers advantages over traditional approaches.

Problem Statement

Introduction

State-of-the-Art Evaluation Strategy

Figure 1: State-of-the-art (a) Time line of contributions, (b) An evaluation

methodology for disparity estimation

Figure 2: Evaluation results by using Middlebury’s methodology

Proposed Evaluation Model

Decision Criteria

Proposed Approach

Evaluation Model

Incommensurability among error measures

Common importance of test bed images and error measures

Results Interpretation

Avoiding ambiguity

Experimental Validation

(a) (b)

Future Work

Among several algorithms of comparable performance which one to

use?

Final Remarks Under the proposed approach, two or more algorithms have a

comparable performance, when their results are not better, neither

worst, since their associated error vectors are incomparable, under a

Pareto dominance criterion.

As innovative aspect, the introduced approach produces significantly

different results in comparison to Middlebury’s evaluation methodology.