Matching Cost Fusion in Dense Depth Recovery for Camera-Array via Global Optimization

This paper proposes a novel method of fusing different matching cost for dense depth map recovery in a global optimization framework. Two simple classical cost functions, NCC and SAD, are combined to make a complementary costs fusion, which is robust against noises and weak radiometric difference. We address complicated difficulties as texture-less region and occlusion in a multi-view energy based global optimization, which is efficiently solved via graph cuts algorithm. We evaluate our cost fusion and optimization algorithm on camera-array captured scenes. The experimental results demonstrate that, our cost fusion get better result than single cost function, and our multi-view optimization gains greatly than stereo method, that means our algorithm is appropriate for camera-array against complex difficulties.