No-reference Sharpness Assessment of In-situ Snapshotted Crystal Images with Multi-Scale Feature Extraction

For evaluating the sharpness of crystal images in-situ detected from a crystallization process, a no-reference sharpness assessment approach is proposed in this paper, by using multi-scale feature extraction in both spatial and spectral domains. The multi-scale re-blurring is firstly adopted on the in-situ collected crystal images to acquire a multi-scale space. Then the sharpness features of crystal images are extracted in the multi-scale space. To extract the gradient features of these images, a local maximum gradient pooling algorithm is subsequently given to procure good performance. Meanwhile, the crystal images are down-sampled multiple times in order to evaluate the discrete cosine transform (DCT) domain entropies of multi-scale resolution. The support vector regression is adopted to map the feature vectors to quality scores. Experimental results on the cooling crystallization experiments of β-form L-glutamic acid well validate the advantage of the proposed sharpness assessment method.