Abstract
In this thesis, we propose a no-reference medical image quality assessment system which measures sharpness and noise to be the image quality index. Based on the ideas of human visual system, the image quality assessment metric is block-based and the JND model, such as contrast weighting and background weighting, is introduced into our proposed quality measures to make the results closer to human perceptual performance. The proposed image quality assessment can work with or without the region of interest information provided by human assessors, which is the most critical region in an image for diagnosis. The proposed medical image quality assessment also has the ability to handle medical image sequences by measuring the gradient correlation on matched guide wire edge blocks. Using this additional information makes our proposed image quality assessment more flexible and practical for diagnostic purpose. The proposed no-reference medical image quality measure is tested by subjective evaluation. By calculating the correlation with MOS, it shows that our proposed method yields good performance and ability to predict the human perceptual medical image quality.