Log-enhanced Salient Edge Depth Map Generation using Deep Neural Networks

Soni, Priyanka and Channappayya, Sumohana (2019) Log-enhanced Salient Edge Depth Map Generation using Deep Neural Networks. Masters thesis, Indian institute of technology Hyderabad.

[img] Text
Thesis_Mtech_CS_5464.pdf - Accepted Version
Restricted to Repository staff only until June 2021.

Download (2MB) | Request a copy


In Full Reference Stereoscopic Image Quality assessment (FRSIQA) Salient Edges with respect to Depth maps (SED) plays an important role. Salient edges are those edges which are prominent to depth perception. Human visual system pays more attention to image saliency compared to depth saliency. The use of disparity maps in FRSIQA have given improved results. However the availability of ground truth disparity maps is difficult. The conventional algorithm of LeSED (Log-enhanced SED maps) generation requires disparity maps along with the left and right images. I propose a deep learning based method to generate LeSED maps. It is an Image to Image translation technique based on Generative Adversarial Networks (GANs) which takes a luminance image and translates it to the LeSED map. Hence we are getting rid of the dependency of ground truth disparity maps as well as a stereo pair. In the first part we generate LeSED maps of stereo pair at the resolution of 256 × 256 and compare the results with conventional algorithm. Our results slightly improved compared to conventional results at this resolution. In the second part we will generate the patched LeSED maps of patched luminace image, as U-net which is a encoder-decoder network takes image size of 256 × 256 resolution. We combine these patched LeSED maps in Fourier domain to go back to the original image resolution. We tested the model on LIVE3DIQAphase1, LIVE3DIQAphase2, IRCCYN and calculated the quality scores. The Quality scores of these LeSED maps have high correlation with DMOS scores, which proves the utility of LeSED maps in FRSIQA

[error in script]
IITH Creators:
IITH CreatorsORCiD
Channappayya, SumohanaUNSPECIFIED
Item Type: Thesis (Masters)
Subjects: Electrical Engineering
Divisions: Department of Electrical Engineering
Depositing User: Team Library
Date Deposited: 14 Jun 2019 11:48
Last Modified: 14 Jun 2019 11:48
URI: http://raiith.iith.ac.in/id/eprint/5464
Publisher URL:
Related URLs:

Actions (login required)

View Item View Item
Statistics for RAIITH ePrint 5464 Statistics for this ePrint Item