Contextual and non contextual evaluation of disparity estimators

Abstract EN

Many stereo disparity computation algorithms have been developed in the literature.

However, the performance evaluation was carried out on different stereo image pairs and only few algorithms used common image test data.

Performance evaluation of stereo algorithms was then subjective (visual).

In order to provide a common ground to evaluate stereo algorithms, it is essential to devise an evaluation strategy.

Moreover, performance evaluation is paramount in order to determine in which conditions an algorithm outperforms another.

This fact certainly guides the choice of the best algorithm with respect to the targeted application.

This paper addresses both non-contextual and contextual evaluation of phase-based stereo algorithms.