Open AccessCompressed Sensing Reconstruction based on Adaptive Scale Parameter using Texture Feature
Author(s) -
D.M.Annie Brighty Christilin*,
M. Safish Mary
Publication year - 2019
Publication title -
international journal of innovative technology and exploring engineering
Language(s) - English
Resource type - Journals
ISSN - 2278-3075
DOI - 10.35940/ijitee.k2246.1081219
Subject(s) - compressed sensing , feature (linguistics) , artificial intelligence , computer science , projection (relational algebra) , iterative reconstruction , random projection , pattern recognition (psychology) , matrix (chemical analysis) , computer vision , sparse matrix , block (permutation group theory) , texture (cosmology) , gaussian , algorithm , image (mathematics) , mathematics , materials science , philosophy , linguistics , physics , geometry , quantum mechanics , composite material
While taking an MRI scan, the patients cannot static for a long time during the motions; the image formation process can create artifacts that may reduce the image quality. The Compressed Sensing (CS) mechanism is employed to reconstruct the original image from the limited data given as the sparse matrix. Hence, CS can be utilized to reduce the acceleration time for an MRI scan considering the patient's health. So the sensing method is implemented by a suitable projection matrix for reconstructing the sparse signals from a few numbers of measurements using Compressed Sensing. The CS guarantees the recovery of the original image with high probability based on random Gaussian projection matrices. However, sparse ternarius projections are more apt for the implementation of hardware. In this article, the proposed deep learning method is employed to obtain a very sparse ternary projection in Compressed Sensing. Compressed Sensing Reconstruction using an adaptive scale parameter based on the texture feature is used to improve the image quality. The two scaling factors αx and αy are assigned to specify the fixed scale for changing the improvement of the image quality. In the parameter using texture feature, the αx and αy are assigned to α as an adaptive scale based on texture feature. In the TACS-SDANN architecture, there are two layers namely the sensing layer which trains the projection matrix and a reconstruction layer which trains for non-linear sparse matrix continuously using Auto-encoder. Experimentally, the scaling factors are calculated on the training data to get the mean PeakSignal-to-Noise Ratio (PSNR) for improving the image quality. Hence a new deep network layer is employed to improve the image quality in this proposed method. Hence the consequence of the proposed method is compared with the SDANN method based on the mean Peak-Signal-to-Noise Ratio (PSNR) to check the image quality. From that comparisons, the TACS-SDANN architecture is proposed to yield a better performance.