Depth data quality evaluation method and device based on graph structure

The invention discloses a depth data quality evaluation method and device based on a graph structure, and the method comprises the steps: obtaining an aligned color and depth RGB-D image, and carrying out the blocking of the RGB-D image, and obtaining a plurality of RGB-D blocks; extracting a feature vector of each RGB-D block, calculating Euclidean distances among all the RGB-D blocks, and connecting each RGB-D block with at least one nearest RGB-D block based on the Euclidean distances among all the RGB-D blocks to generate a simple graph corresponding to an adjacent matrix; and performing graph convolution on the feature vector of each RGB-D block by using the adjacent matrix to generate a one-dimensional vector, and generating a local evaluation score of the RGB-D image by using the one-dimensional vector. Therefore, the problems that in the prior art, under the condition that a reference depth map is missing, an accurate quality score cannot be calculated for a distorted depth map, only simple features a