Multi-Directional Feature Prediction Prior and Enhanced Non-Local Total Variation Model for Single Image Super Resolution

PROJECT TITLE :

Enhanced Non-Local Total Variation Model and Multi-Directional Feature Prediction Prior for Single Image Super Resolution

ABSTRACT:

Single image super-resolution (SISR) approaches are widely recognised to play a crucial role in recovering the missing high frequencies in an input low-resolution image. Image priors are required to regularise the solution spaces and generate the appropriate high-resolution image because SISR is extremely ill-conditioned. Based on the improved non-local similarity modelling and multi-directional feature prediction in this paper, we have developed an effective SISR framework (ENLTV-MDFP). The suggested ENLTV-MDFP method benefits from the synergistic qualities of both reconstruction-based and learning-based SISR approaches because both priors are used. It is characterised by the fading kernel and stable group similarity reliability techniques for the non-local similarity-based modelled prior (ENLTV). The deep convolutional neural network is used to learn the prior [multi-directional feature prediction prior (MDFP)]. Edges and visual artefacts can be improved by using the modelled prior, while details can be hallucinated using the learnt prior. A combined SR cost function is proposed by combining these two complimentary priors in the MAP framework. Finally, using the split Bregman iteration technique, the combined SR issue is solved. The suggested ENLTV-MDFP technique beats many current state-of-the-art algorithms in terms of both visual and quantitative performance.

Did you like this research project?

To get this research project Guidelines, Training and Code... Click Here