High Quality Depth Map Reconstruction from Sparse Samples using Alternating Direction Method of Multipliers

Roshna V N; Abdullah Gubbi

CALL FOR PAPERS : Dec-2023

Submission Last Date

25-Dec-23

Submit Manuscript Online

FOR AUTHORS

FOR REVIEWERS

ARCHIEVES

DOWNLOADS

Open Access

High Quality Depth Map Reconstruction from Sparse Samples using Alternating Direction Method of Multipliers

Author(s):

Roshna V N , P A College of engineering Mangalore; Abdullah Gubbi, P A College of engineering Mangalore

Keywords:

ADMM, PCA, Depth estimation, wavelet contourlet dictionaries, two staged random sampling

Abstract:

The rapid progress in 3D technology has enhanced whole the systems to be focused on digital world. Digital analysis mainly include sampling, encoding, decoding the processed data. Depth estimation is one of the interesting phases where we have to adopt new methods that will improve the depth map. Existing methods like hardware and computational procedures leads to high expense as well as deprived depth accuracy. The proposed method highlights how the depth image is reconstructed precisely by adopting proficient computational techniques. This is mainly divided into three sections. First part deals with representation of input depth image which is encoded by using dictionaries. It is preferable to use combination of wavelet contourlet dictionaries rather than single dictionary like wavelet which gives enhanced performance with high peak to signal ratio. Second section propose effective algorithm for reconstruction of depth image named as alternating direction method of multipliers (ADMM). This algorithm is in co-operated with the combined dictionaries and alone in order to fix the best dictionary. In the third section sampling technique with high peak to signal ratio is chosen. Two staged random sampling technique which is improved by principal component analysis (PCA) is being used to pick relevant sampling points for the precise depth reconstruction. Improved depth map with least mean square error is obtained by carefully choosing the dictionary, algorithm and sampling technique.