Browsing by Author "Nair, M.S."

Now showing 1 - 5 of 5

An empirical study of the impact of masks on face recognition
(Elsevier Ltd, 2022) Jeevan, G.; Zacharias, G.C.; Nair, M.S.; Rajan, J.
Face recognition has a wide range of applications like video surveillance, security, access control, etc. Over the past decade, the field of face recognition has matured and grown at par with the latest advancements in technology, particularly deep learning. Convolution Neural Networks have surpassed human accuracy in Face Recognition on popular evaluation tests such as LFW. However, most existing models evaluate their performance with an assumption of the availability of full facial information. The COVID-19 pandemic has laid forth challenges to this assumption, and to the performance of existing methods and leading-edge algorithms in the field of face recognition. This is in the wake of an explosive increase in the number of people wearing face masks. The reduced amount of facial information available to a recognition system from a masked face impacts their discrimination ability. In this context, we design and conduct a series of experiments comparing the masked face recognition performances of CNN architectures available in literature and exploring possible alterations in loss functions, architectures, and training methods that can enable existing methods to fully extract and leverage the limited facial information available in a masked face. We evaluate existing CNN-based face recognition systems for their performance against datasets composed entirely of masked faces, in contrast to the existing standard evaluations where masked or occluded faces are a rare occurrence. The study also presents evidence denoting an increased impact of network depth on performance compared to standard face recognition. Our observations indicate that substantial performance gains can be achieved by the introduction of masked faces in the training set. The study also inferred that various parameter settings determined suitable for standard face recognition are not ideal for masked face recognition. Through empirical analysis we derived new value recommendations for these parameters and settings. © 2021 Elsevier Ltd
Guided SAR image despeckling with probabilistic non local weights
(Elsevier Ltd, 2017) Gokul, J.; Nair, M.S.; Rajan, J.
SAR images are generally corrupted by granular disturbances called speckle, which makes visual analysis and detail extraction a difficult task. Non Local despeckling techniques with probabilistic similarity has been a recent trend in SAR despeckling. To achieve effective speckle suppression without compromising detail preservation, we propose an improvement for the existing Generalized Guided Filter with Bayesian Non-Local Means (GGF-BNLM) method. The proposed method (Guided SAR Image Despeckling with Probabilistic Non Local Weights) replaces parametric constants based on heuristics in GGF-BNLM method with dynamically derived values based on the image statistics for weight computation. Proposed changes make GGF-BNLM method adaptive and as a result, significant improvement is achieved in terms of performance. Experimental analysis on SAR images shows excellent speckle reduction without compromising feature preservation when compared to GGF-BNLM method. Results are also compared with other state-of-the-art and classic SAR depseckling techniques to demonstrate the effectiveness of the proposed method. © 2017 Elsevier Ltd
Iterative bilateral filter for Rician noise reduction in MR images
(Springer London, 2015) Riji, R.; Rajan, J.; Sijbers, J.; Nair, M.S.
Noise removal from magnetic resonance images is important for further processing and visual analysis. Bilateral filter is known for its effectiveness in edge-preserved image denoising. In this paper, an iterative bilateral filter for filtering the Rician noise in the magnitude magnetic resonance images is proposed. The proposed iterative bilateral filter improves the denoising efficiency, preserves the fine structures and also reduces the bias due to Rician noise. The visual and diagnostic quality of the image is well preserved. The quantitative analysis based on the standard metrics like peak signal-to-noise ratio and mean structural similarity index matrix shows that the proposed method performs better than the other recently proposed denoising methods for MRI. © 2014, Springer-Verlag London.
Single image super resolution from compressive samples using two level sparsity based reconstruction
(2015) Nath, A.G.; Nair, M.S.; Rajan, J.
Super Resolution based on Compressed Sensing (CS) considers low resolution (LR) image patch as the compressive samples of its high resolution (HR) patch. Compressed sensing based image acquisition systems acquire less number of random linear measurements without first collecting all the pixel values. But using these compressive measurements directly to reconstruct the image causes quality issues. In this paper an image super-resolution method with two level sparsity based reconstruction via patch based image interpolation and dictionary learning is proposed. The first level reconstruction generates a low resolution image from random samples and the interpolation scheme used in this algorithm reduces the HR-LR patch coherency due to neighborhood issue which is a major drawback of single image super resolution algorithms. The dictionary based reconstruction phase generates the high resolution image from the low resolution output of the first level reconstruction phase. The experimental results proved that the proposed two level reconstruction scheme recovers more details of the image and yields improved results from very few samples (around 35-45%) than the state-of-the-art algorithms which uses low resolution image itself as input. The results are compared by considering both PSNR values and visual perception. � 2015 The Authors.
Single image super resolution from compressive samples using two level sparsity based reconstruction
(Elsevier B.V., 2015) Nath, A.G.; Nair, M.S.; Rajan, J.
Super Resolution based on Compressed Sensing (CS) considers low resolution (LR) image patch as the compressive samples of its high resolution (HR) patch. Compressed sensing based image acquisition systems acquire less number of random linear measurements without first collecting all the pixel values. But using these compressive measurements directly to reconstruct the image causes quality issues. In this paper an image super-resolution method with two level sparsity based reconstruction via patch based image interpolation and dictionary learning is proposed. The first level reconstruction generates a low resolution image from random samples and the interpolation scheme used in this algorithm reduces the HR-LR patch coherency due to neighborhood issue which is a major drawback of single image super resolution algorithms. The dictionary based reconstruction phase generates the high resolution image from the low resolution output of the first level reconstruction phase. The experimental results proved that the proposed two level reconstruction scheme recovers more details of the image and yields improved results from very few samples (around 35-45%) than the state-of-the-art algorithms which uses low resolution image itself as input. The results are compared by considering both PSNR values and visual perception. Â© 2015 The Authors.