Δευτέρα 5 Ιουνίου 2017

Blind multi-image super-resolution based on combination of ANN learning and non-subsampled Contourlet directional image representation

Abstract

In image super-resolution technique, it is contradictory to keep the edge characteristics of the image while de-noising. In order to solve the above problem, we propose a blind multi-image super-resolution algorithm which is adaptive to the image content and does not need the fuzzy conditions of generating the low-resolution images. The initially estimated high-resolution image \(\left( \hat{H}\right) \) is firstly gotten through the traditional reconstruction algorithm. Afterward, the estimation differences of low-resolution images are utilized as the input of artificial neural network (ANN), and the band-pass directional sub-bands of non-subsampled Contourlet transform (NSCT) of the lost high-frequency components are outputted in ANN. With the inverse NSCT, we can get the estimated lost high-frequency components. Finally, the estimated lost high-frequency components and the adaptive weighted matrix generated from the image content are multiplied before being added to \(\hat{H}\) . Experimental results show that the high-resolution images obtained through the proposed method can achieve favorable subjective and objective quality for different image contents. Meantime, it is superior to some of the state-of-the-art classical methods in terms of the performance.



http://ift.tt/2qTfRqn

Δεν υπάρχουν σχόλια:

Δημοσίευση σχολίου

Δημοφιλείς αναρτήσεις