The invention discloses a no-reference stereo image quality evaluation method based on dictionary learning and machine learning. The method comprises the following steps of: firstly, performing log-Gabor filtering for left and right viewpoint images, obtaining respective amplitude and phase information, then, performing local binarization operation for the amplitude and the phase information, and obtaining a local binarization mode feature image of the left and right viewpoint images; secondly, using a binocular energy model to fuse the amplitude and the phase information of the left and right viewpoint images, obtaining binocular energy information, and acquiring a local binarization mode feature image of the binocular energy information; then, using a coordination representation algorithm to perform dictionary learning for the local binarization mode feature images of the left and right viewpoint images and the binocular energy information, obtaining binocular visual perception sparse feature information, and finally, obtaining an objective quality evaluation predicted value of a to-be-evaluated distorted stereo image. The method has the advantages of being capable of fully considering stereo visual perception characteristics, and being capable of effectively improving correlation between objective evaluation result and subjective perception.