Superpixel-based Structural Similarity Metric for Image Fusion Quality Evaluation

• Published in: Sensing and imaging Vol. 22; no. 1
• Main Authors: Wang, Eryan, Yang, Bin, Pang, Lihui
• Format: Journal Article
• Language: English
• Published: New York Springer US 01-12-2021; Springer Nature B.V
• Subjects: Computer vision; Electrical Engineering; Engineering; Feature extraction; Image processing; Image quality; Imaging; Microwaves; Original Paper; Quality assessment; Radiology; RF and Optical Engineering; Similarity; Stability analysis; Superpixel Segmentation; Image Fusion Quality Assessment; Human Visual System; Adaptive Superpixel; Image Feature; Structural Similarity Metric
• ISSN: 1557-2064; 1557-2072
• DOI: 10.1007/s11220-021-00339-1

Image fusion refers to integrate multiple images of the same scene into a high-quality fused image. Universal quality evaluation for fused image is one of the urgent problems in the field of image fusion. Typically, local features extracted from rectangular blocks of the fused images are used to achieve objective evaluation. However, the fixed shape of image block is neither suitable for the natural attributes of an image, nor for the perceptual characteristics of human visual system. To deal with the problem, a superpixel-based structural similarity metric for image fusion quality evaluation is proposed in this paper. The image features extracted from adaptive superpixels are used to calculate the structural similarity between the corresponding superpixels. Then all local structural similarity indicators are weighted and averaged according to their significance to obtain the final evaluation score. Several classical image fusion quality evaluation metrics are used for comparative experimental analysis. A series of experimental results show that the stability of the proposed quality evaluation index is about 10
−6
orders of magnitude, whose accuracy and performance are more advantageous than the latest evaluation index. Meanwhile, the evaluation results obtained by the proposed metric is closer to the human visual evaluation results.