Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
17 results
Search Results
Item Frontal Gait Recognition based on Hierarchical Centroid Shape Descriptor and Similarity Measurement(Institute of Electrical and Electronics Engineers Inc., 2019) Anusha, R.; Jaidhar, C.D.Gait recognition is an expanding stream in biometrics, intended to recognize individuals through the investigation of their walking pattern. This pattern is obtained from a distance, without the active participation of the people. One of the difficulties of the appearance-based gait approach is to enhance the performance of frontal gait recognition, as it carries less spatial and temporal data when compared with other view variations. As a result, to increase the performance of the frontal gait recognition, this paper presents a method which uses two-step procedure; the Hierarchical centroid Shape descriptor (HCSD) and the similarity measurement. The proposed method was assessed on the broadly used CASIA A, CASIA B, and CMU MoBo gait databases. The experimental outcomes showed that the proposed method gave promising results and outperforms certain state-of-the-art methods in terms of recognition performance. © 2019 IEEE.Item An Approach to Speed Invariant Gait Analysis for Human Recognition using Mutual Information(Institute of Electrical and Electronics Engineers Inc., 2019) Anusha, R.; Jaidhar, C.D.Gait is a biometric characteristic that facilitates the identification of individuals with low-resolution images. This aspect intensifies its utility in many human detection applications. However, there are many challenges that adversely affect the gait recognition performance. They are caused by the impact of various covariate aspects such as, changes in clothing and carrying conditions, walking speed, walking surface conditions, view variations, and so on. This paper proposes an effective approach for speed-invariant gait recognition system. This approach uses the Region of Interest (ROI) extracted from Gait Energy Image (GEI) to classify a probe sample into a gallery sample. The mutual information obtained from a probe and gallery sample, followed by their classification capture the spatial dynamics of GEI efficiently to improve the gait recognition performance. Further, the proposed method is evaluated on CASIA C and OU-ISIR Treadmill A gait databases. Experimental results demonstrate the capability of the proposed approach in comparison with the existing gait recognition methods. © 2019 IEEE.Item Gaussian Filtered Gait Energy Template and Centroid Corner Distance Features for Human Gait Recognition(Institute of Electrical and Electronics Engineers Inc., 2019) Anusha, R.; Jaidhar, C.D.One of the convincing and latest biometric systems is gait recognition because of its ability to unobtrusively identify an individual at a distance and with low-resolution images. This study proposes an efficient method to enhance the performance of the gait detection system. The gait silhouette images are initially processed with two gait portrayal methods as the feature resources: Gait Energy Image (GEI) and Gaussian Filtered-Gait Energy Image (GF-GEI). Further, an effort has been made to present a statistical shape examination method, which is established on GF-GEI, and it is divided into six independent horizontal segments. The centroid corner distance features obtained from these horizontal segments forms the feature vector of the image. The proposed method is assessed on the widely used CASIA A, CASIA B, and OU-ISIR D gait datasets. The empirical results illustrate that the performance of the proposed approach is promising and surpasses some state-of-the-art gait identification methods recorded in literature. © 2019 IEEE.Item Improved Robustness of EMG Pattern Recognition for Transradial Amputees with EMG Features Against Force Level Variations(Institute of Electrical and Electronics Engineers Inc., 2023) Powar, O.S.; Chemmangat, K.Feature extraction is an essential process for removing the unwanted part and interference of the Electromyography (EMG) signal, and to extract the useful information hidden in it. Inorder to obtain high performance of Myoelectric Control (MEC), the choice of features plays an important role. The studies carried out earlier to overcome force level variation have used features which are redundant, affecting the robustness and the classification performance. This study's main objective is to assess a database's performance consisting of nine upper limb amputee subjects with EMG data recorded at three different force levels when six motions were classified using twenty different time domain features that are frequently found in the literature. Training is carried out at one force level, and the other two unknown force levels are used for testing. Out of the twenty features, the one that is the most stable is displayed for each force level. The results show that root mean square (RMS) feature outperformed other features for training at low and medium force levels, and Wilson amplitude (WAMP) feature for training at a high force level, when compared with the most widely used linear discriminant analysis (LDA) classifier. The average classification accuracy for the nine amputee subjects trained with the RMS feature at low and medium force levels was 42% and 51.78% percent, respectively. For high force level, when trained using WAMP feature, an accuracy of 46.78% has been obtained. The features are verified using histogram plots. This study will help select those features which are not important for robust classification of hand movements. © 2023 IEEE.Item Spatial Dynamics for Identification of Individuals through Gait and Other Locomotion Activities(Institute of Electrical and Electronics Engineers Inc., 2024) Anusha, R.; Sanshi, S.Gait, the pattern of walking, has been extensively studied and various methods have been developed to use it as a biometric for individual recognition. Despite this, the potential to identify individuals through running videos has not been thoroughly explored. The paper introduces a novel method that expands the feature-based approach for identifying individuals based on their running style. This work focuses on extracting the mutual information and location specific metric from the key gait poses of subjects in the testing and training datasets. Later on, the assignment of a testing sample to the training sample is accomplished using the proposed classification method. The experiments are conducted on KTH, OU-ISIR A, and Weizmann database. The efficiency of this method is demonstrated by the obtained experimental results. © 2024 IEEE.Item Automatic seizure detection by modified line length and Mahalanobis distance function(Elsevier Ltd, 2018) Pathak, A.; Ramesh, A.; Mitra, A.; Majumdar, K.Automatic seizure detection with high accuracy and in linear time has profound implications on therapeutic intervention mechanisms. In this work taking into account 12 popular seizure detection algorithms we have shown that line length is one feature that is extractable in linear time from EEG signals and capable of automatic seizure onset detection with highest accuracy among linear time extractable features. Also line length is less prone to give false positives. The detection accuracy has been ascertained by ROC curve analysis on Freiburg Seizure Prediction Project data containing intracranial EEG recordings of 87 seizures from 21 patients with sufficient interictal signals. Next, we have modified the classical line length feature extraction algorithm to improve its accuracy without any additional computational burden. Finally, we have applied both classical line length (LL) and modified line length (MLL) on all focal channels and detected seizures on multidimensional focal channel signals by Mahalanobis distance function (MDF). Both detected 73 out of 87 seizures. Area under the ROC curve (AUC), detection delay and false positive for LL and MLL are 0.951, 11.903 s, 0.201/h and 0.954, 11.698 s, 0.198/h respectively. Since LL has already been incorporated into an FDA approved commercially available closed loop intervention system, even this minute improvement may have significant healthcare implications. © 2018 Elsevier LtdItem Automatic detection and localization of Focal Cortical Dysplasia lesions in MRI using fully convolutional neural network(Elsevier Ltd, 2019) Bijay Dev, K.M.; Pawan, P.S.; Niyas, S.; Vinayagamani, S.; Kesavadas, C.; Rajan, J.Focal cortical dysplasia (FCD) is the leading cause of drug-resistant epilepsy in both children and adults. At present, the only therapeutic approach in patients with drug-resistant epilepsy is surgery. Hence, the quantification of FCD via non-invasive imaging techniques helps physicians to decide on surgical interventions. The properties like non-invasiveness and capability to produce high-resolution images makes magnetic resonance imaging an ideal tool for detecting the FCD to an extent. The FCD lesions vary in size, shape, and location for different patients and make the manual detection time consuming and sensitive to the experience of the observer. Automatic segmentation of FCD lesions is challenging due to the difference in signal strength in images acquired with different machines, noise, and other kinds of distortions such as motion artifacts. Most of the methods proposed in the literature use conventional machine learning and image processing techniques in which their accuracy relies on the trained features. Hence, feature extraction should be done more precisely which requires human expertise. The ability to learn the appropriate features/representations from the training data without any human interventions makes the convolutional neural network (CNN) the suitable method for addressing these drawbacks. As far as we are aware, this work is the first one to use a CNN based model to solve the aforementioned problem using only MRI FLAIR images. We customized the popular U-Net architecture and trained the proposed model from scratch (using MRI images acquired with 1.5T and 3T scanners). FCD detection rate (recall) of the proposed model is 82.5 (33/40 patients detected correctly). © 2019Item Reducing the effect of wrist variation on pattern recognition of Myoelectric Hand Prostheses Control through Dynamic Time Warping(Elsevier Ltd, 2020) Powar, O.S.; Chemmangat, K.For upper limb prostheses, research carried out earlier mainly focused on increasing the classification accuracy of the hand movements; but there exist a little work done on factors affecting it in real-time control such as wrist variation. Amputees with functional wrist use their prostheses in multiple wrist positions. Since the Electromyography (EMG) data is taken while the subject is performing the motion in different wrist position, it can degrade the performance of the Pattern Recognition (PR) system. In this work, a wrist independent PR scheme has been developed. In this regard, Dynamic Time Warping (DTW) is used to overcome the effects due to wrist variation. The performance of the DTW scheme as a PR system is validated using two training methods; with classification accuracy as a performance measure on data taken from the database of ten intact subjects for six hand motions carried out at three different wrist orientations. On the database, an average classification accuracy of about 93.3% was obtained while trained using EMG data from all possible wrist positions. The effectiveness of the method is demonstrated in terms of classification accuracy and processing time when compared with the Time-domain power spectral descriptors (TD-PSD) method which outperformed other methods in the literature for reducing the impact of wrist variation on EMG based PR. The results show that the DTW can be a computationally cheap and accurate PR system for real-time hand movement classification. © 2019 Elsevier LtdItem NucleiSegNet: Robust deep learning architecture for the nuclei segmentation of liver cancer histopathology images(Elsevier Ltd, 2021) Lal, S.; Das, D.; Alabhya, K.; Kanfade, A.; Kumar, A.; Kini, J.R.The nuclei segmentation of hematoxylin and eosin (H&E) stained histopathology images is an important prerequisite in designing a computer-aided diagnostics (CAD) system for cancer diagnosis and prognosis. Automated nuclei segmentation methods enable the qualitative and quantitative analysis of tens of thousands of nuclei within H&E stained histopathology images. However, a major challenge during nuclei segmentation is the segmentation of variable sized, touching nuclei. To address this challenge, we present NucleiSegNet - a robust deep learning network architecture for the nuclei segmentation of H&E stained liver cancer histopathology images. Our proposed architecture includes three blocks: a robust residual block, a bottleneck block, and an attention decoder block. The robust residual block is a newly proposed block for the efficient extraction of high-level semantic maps. The attention decoder block uses a new attention mechanism for efficient object localization, and it improves the proposed architecture's performance by reducing false positives. When applied to nuclei segmentation tasks, the proposed deep-learning architecture yielded superior results compared to state-of-the-art nuclei segmentation methods. We applied our proposed deep learning architecture for nuclei segmentation to a set of H&E stained histopathology images from two datasets, and our comprehensive results show that our proposed architecture outperforms state-of-the-art methods. As part of this work, we also introduced a new liver dataset (KMC liver dataset) of H&E stained liver cancer histopathology image tiles, containing 80 images with annotated nuclei procured from Kasturba Medical College (KMC), Mangalore, Manipal Academy of Higher Education (MAHE), Manipal, Karnataka, India. The proposed model's source code is available at https://github.com/shyamfec/NucleiSegNet. © 2020 Elsevier LtdItem An effective feature extraction with deep neural network architecture for protein-secondary-structure prediction(Springer, 2021) Jayasimha, A.; Mudambi, R.; Pavan, P.; Lokaksha, B.M.; Bankapur, S.; Patil, N.With the increased importance of proteins in day-to-day life, it is imperative to know the protein functions. Deciphering protein structure elucidates protein functions. Experimental approaches for protein-structure analysis are expensive and time-consuming, and require high dexterity. Thus, finding a viable computational approach is vital. Due to the high complexity of predicting protein structure (tertiary structure) directly, research in this field aims at the protein-secondary-structure prediction which is directly related to its tertiary structure. This research aims at exploring a plethora of features, namely position-specific scoring matrices, hidden Markov model alignment matrices, and physicochemical properties, that carry rich information required to predict the secondary structure. Furthermore, it aims at exploring a suitable combination of the features which could capture diverse information about the protein secondary structure. Finally, a cascaded convolutional neural network and bidirectional long short-term memory architecture is fit on the models, and two evaluation metrics, namely, Q8 score and segment overlap score, are benchmarked on various datasets. Our proposed model trained on data of CB6133 dataset and tested on CB513 dataset beats the benchmark models by a minimum of 2.9%. © 2021, The Author(s), under exclusive licence to Springer-Verlag GmbH Austria, part of Springer Nature.