Fast stereo matching algorithm based on adaptive window

Abstract

Depth generation is the key technology in the 3DTV system. Due to the demands for accuracy and real-time execution, local methods are more suitable, but the challenge is to decide an appropriate window for each pixel. We propose a novel local method, which constructs the support windows adaptively only based on the intensity, and the process has no relation with disparities. This is the main difference between this algorithm and others. Furthermore, we take integral image technique to accelerate cost aggregation over support windows. To demonstrate the effectiveness of the algorithm, it has been tested on the benchmark Middlebury stereo database. The results show that our algorithm has comparable efficiency, and hence it is promising for practical application.