No-reference blur image quality measure based on multiplicative multiresolution decomposition

• A multiplicative multi-resolution decomposition MMD is used for blur analysis.
• Indexes are extracted by image-content analysis through MMD coefficients.
• A quality blur measure is proposed by using a set of thresholds (direct MMD-BQ-1 and iterative MMD-BQ-2 ).
• A quality blur measure is proposed by using a learning process MMD-BQSVM.

A new approach for analyzing the blur effect on real images is proposed. This approach is based on the Multiplicative Multi-resolution Decomposition MMD. From MMD image-content analysis, a blind image quality measure dedicated to blur is then derived. The proposed measure has been applied on Gaussian-blurred and JPEG2000-compressed images from the LIVE, TID and IVC databases. The performance of the proposed measure is evaluated and compared with some referenced image quality metrics. The experimental results measured in terms of correlation with the subjective assessment of the images, demonstrate the efficiency of the proposed image quality measure in predicting the amount of blur.