Treffer: Image quality assessment and multi-focus image fusion

Ph.D. ; In the last decades, deep learning develops fast and achieves great success in many image processing tasks. The achievements prove that deep learning is a powerful tool. Therefore, I tried to adopt deep learning to solve two image processing tasks, image quality assessment and multi-focus image fusion. ; Image Quality Assessment (IQA) targets on objectively estimating the visual quality of an image, where the quality score is expected to be well correlated well with the human subjective visual quality score. Among all IQA metrics, the no-reference non-distortion specific ones are most challenging. They aim at accurately evaluating images distorted by any distortion types without the help of the reference image. Though different distortion types may lead to the different influence on Human Visual System(HVS), we focus on proposing non-distortion-specific models. During this process, the distortion information is a crucial clue to be utilized. ; An image quality evaluator, i.e. Deep Learning based Blind Image Quality Index(DL-BIQI), was first explored and designed to make full use of the distortion information. In DL-BIQI, several models are designed for several specific distortion types. Meanwhile, another deep classification model is proposed to estimate the presence of a set of distortions in the testing image. The final visual quality is obtained by a probability-weighted summation model. Experiments were conducted on the LIVE dataset to evaluate the effectiveness of DL-BIQA. The performance of the DL-BIQI achieves 0.951 for Spearman Rank-Order Correlation Coefficient (SROCC). It outper forms many state-of-the-art methods for comparison. Besides, the proposed distortion type classification model achieves 93.7% accuracy on the LIVE dataset. ; The success of DL-BIQI proves the effectiveness of distortion information in estimating visual quality. Inspired by this, another learning based IQA framework, Visual Importance and Distortion Guided Image Quality Assessment method (VIDGIQA), is proposed where the ...