Matching Cost Calculation Model Based on Multi-Scale Convolutional Neural Network

Aiming at the problem that the pixel matching cost is difficult to accurately calculate in complex images, a matching cost calculation model based on multi-scale convolutional neural network is proposed in this paper. The proposed calculation model optimizes the existing model based on feature fusion idea. This model improves the feature extraction ability by extracting and fusing different scale feature information. The experiment results show that the pixel matching accuracy of the matching cost calculation model based on multi-scale convolutional neural network is 8% higher than the existing matching cost calculation model.