Loading

Salient Object Detection on Matrix Disintegration using Convolutional Neural Network
Hemraj Singh1, Maheep Singh2

1Hemraj Singh, Sumandeep Vidyapeeth, Taluka Waghodia (Gujarat), India.

2Maheep Singh, Sumandeep Vidyapeeth, Taluka Waghodia (Gujarat), India.

Manuscript received on 02 February 2021 | Revised Manuscript received on 17 February 2021 | Manuscript Accepted on 15 March 2021 | Manuscript published on 30 March 2021 | PP: 8-14 | Volume-1 Issue-1, March 2021 | Retrieval Number: A1003031121/2021©LSP

Open Access | Ethics and Policies | Indexing and Abstracting
© The Authors. Published by Lattice Science Publication (LSP). This is an open access article under the CC-BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/)

Abstract: In this work, initially low-rank matrix is used for Salient object detection (SOD) and generating saliency. The Saliency map obtained using low rank matrix is then passed to the convolutional neural network (CNN) to obtain the final saliency map. Salient object has been recognized by sparse matrix and is detected using CNN, but still there remains two problems to be resolved. First, the elements in the sparse matrix which are commonly self-sufficient, avoids spatial pattern of image regions. Second, if the entire low-rank matrix and sparse matrix have been fairly well-founded, then a likeliness of the salient object and background are convoluted. To solve these problems we have proposed a novel model with two conditions: (1) We have used tree-structure sparsity breeding formalization which captures image structure and revamp to the object which has akin saliency, (2) Laplacian formalization has been used for filling the gaps in the images that contains multiple salient objects. Overall, high level priors have been used to handle matrix decomposition and to increase the salient object detection capacity, and to detect the salient object a CNN network has been designed that handles single, multiple and complex scene of images and gives the better result as compared to state-of-the-art models.

Keywords: Sparse matrix, convolutional neural network, Low-rank matrix, subspace, structured skimp.