Finding the Largest Unambiguous Component of Stereo Matching

Abstract

Abstract. Stereo matching is an ill-posed problem for at least two principal reasons: (1) because of the random nature of match similarity measure and (2) because of structural ambiguity due to repetitive patterns. Both ambiguities require the problem to be posed in the regularization framework. Continuity is a natural choice for a prior model. But this model may fail in low signal-to-noise ratio regions. The resulting artefacts may then completely spoil the subsequent visual task. A question arises whether one could (1) find the unambiguous component of matching and, simultaneously, (2) identify the ambiguous component of the solution and then, optionally, (3) regularize the task for the ambiguous component only. Some authors have already taken this view. In this paper we define a new stability property which is a condition a set of matches must satisfy to be considered unambiguous at a given confidence level. It turns out that for a given matching problem this set is (1) unique and (2) it is already a matching. We give a fast algorithm that is able to find the largest stable matching. The algorithm is then used to show on real scenes that the unambiguous component is quite dense (10-80%) and error-free (total error rate of 0.3-1.4%), both depending on the confidence level chosen.

Dense Stereomatching Algorithm Performance for View Prediction and Structure Reconstruction

“…We have chosen four area-based algorithms employing strong and weak continuity priors: The former represented by MAP matching via graph cuts [6] (GC) and ML matching via dynamic programming [2] (DP), and the later by Confidently Stable Matching [10] (CSM) and Winner-Take-All [11] (WTA) algorithms, respectively.…”

“…In the DP and GC the parameters are set to minimize the FR error under the middle texture contrast level (the T-level). The parameters of CSM represent the matching confidence level and thus directly determine the MIR over a very wide range [10] (low MIR implies high FNR and vice versa). The parameters of CSM are set so that the MIR becomes equal to that of the GC under the maximum contrast.…”

Robust correspondence recognition for computer vision

“…A faster algorithm for stereovision is described in [Sar02] and a modification that produced the results in Figs. 2 and 3 is described in [KS03].…”

“…Of course one should discard as little as possible but prior knowledge useful for such disposal is hard if not impossible to obtain. With the exception of [GY05,Sar02], none of the known algorithms models occlusion properly and exhibits the ability to reject part of input data that is required here.…”

“…A robust behavior with respect to partial invalidity of the prior model is demonstrated in Fig. 2(b): The algorithm also runs per image lines and assumes ordering (the result is a stable solution and the algorithm is graph kernel which described in [Sar02,KS03] and later in this paper). Note there are no streaks.…”