Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
8 results
Search Results
Item Cell Segmentation by Modified U-Net Architecture for Biomedical Images(Institute of Electrical and Electronics Engineers Inc., 2020) Kumar, C.A.; Kumar, M.T.N.; Narasimhadhan, A.V.Biomedical image segmentation is one of the main and fast growing field in medical image processing domain. Deep neural networks is one of the popular field used for image segmentation. Convolutional neural networks(CNNs) in deep neural networks have shown good performance for biomedical image segmentation. However, a strong notion exists that large number of annotated images are required for training of CNNs. Therefore, in this paper we have come up with a modified U-Net architecture for limited number of annotated data with an intersection over union score of 92.54%. The architecture uses rectified-adam optimizer(advanced version of adam) for minimizing the loss function which helps us to come close to global optima. We have also compared the performance of various optimizers on the proposed network. © 2020 IEEE.Item 3D AttU-NET for Brain Tumor Segmentation with a Novel Loss Function(Institute of Electrical and Electronics Engineers Inc., 2023) Roy, R.; Annappa, B.; Dodia, S.In the United States of America (USA), every year 150,000 patients are registered with a secondary brain tumor that is not generated in the brain. This necessitates the need for early brain tumor detection, which in turn will help patients to live longer. For clinical evaluation and treatment, precise segmentation of brain tumors in MRI images is required. This process can be aided by machine learning and efficient image processing, but manual imaging can be time-consuming. In this study, we aim to develop an 3D automated segmentation algorithm with a novel loss function. A 3D attention UNET CNN model was trained using the novel loss function, which was calculated by taking the weighted average of dice loss and focal loss to overcome the class imbalance. Results show the enhancement in the segmentation performance of attention UNET model with an average increase of 5% in the Dice coefficient for all three classes. However, the model's performance was not as strong for enhanced and core tumors. Further research may be needed to optimize performance in these areas. . © 2023 IEEE.Item A Survey on Semantic Segmentation Models for Underwater Images(Springer, 2023) Anand, S.K.; Kumar, P.V.; Saji, R.; Gadagkar, A.V.; Chandavarkar, B.R.Semantic segmentation remains a key research field in modern day computer vision and has been used in a myriad of applications across various fields. It can be extremely beneficial in the study of underwater scenes. Various underwater applications, such as unmanned explorations and autonomous underwater vehicles, require accurate object classification and detection to allow the probes to avoid malicious objects. However, the models that work well for terrestrial images rarely work just as well for underwater images. This is because underwater images suffer from high blue light intensity as well as other ill effects such as poor lighting and contrast. This can be fixed using preprocessing techniques to manually improve the image characteristics. Trying to improve the model to account for bad image quality is not a great method as the model may misidentify noise as an image characteristic. In this chapter, 6 different deep learning semantic segmentation models—SegNet, Pyramid Scene Parsing Network (PSP-Net), U-Net, DNN-VGG (Deep Neural Network-VGG), DeepLabv3+, and SUIM-Net—are explored. Their architectures, technical aspects with respect to underwater images, advantages, and disadvantages are all investigated. © 2023, The Author(s), under exclusive license to Springer Nature Switzerland AG.Item 3D-Conditional Generative Adversarial Networks for Brain Tumour Segmentation(Institute of Electrical and Electronics Engineers Inc., 2024) Magar, P.K.; Naik, D.Gliomas, the most common primary brain tumor, exhibit significant heterogeneity in their prognosis, aggressiveness, and histological composition, encompassing areas such as necrotic cores, enhancing and non-enhancing tumor cores, and peritumoral edema. While multimodal MRI is invaluable for brain tumor detection, precise tumor segmentation remains challenging. To overcome this, a novel 3D volume-to-volume GAN, termed the 3D-Conditional Generative Adversarial Network (3D-cGAN), was developed for the brain tumor segmentation, leveraging data from the 2020 BraTS Challenge. This model utilizes multi channel 3D MRI images to accurately segment core, whole, and enhancing tumor regions. Employing a batch size of 4 and an alpha value of 2, the model demonstrates remarkable accuracy on the BraTS 2020 dataset, achieving a Dice score of 0.8286 and IoU score of 0.7111. © 2024 IEEE.Item Automated Segmentation of COVID-19 Infected Lungs via Modified U-Net Model(Institute of Electrical and Electronics Engineers Inc., 2024) Kumar, S.; Bhowmik, B.The COVID-19 pandemic has led to significant outbreaks in more than 220 countries worldwide, profoundly impacting the public health and lives. As of February 2024, over 774 million cases have been reported, with more than 7,035,337 deaths recorded. Therefore, there is a significant need for automated image segmentation to serve as clinical decision support. This paper presents a novel automated segmentation framework that dynamically generates distinct and randomized image patches for training using preprocessing techniques and extensive data augmentation. The proposed architecture employs a semantic segmentation approach, ensuring accuracy despite limited data availability. Experimental assessment comprises a visual inspection of the predicted segmentation outcomes. Quantitative evaluation of segmentation includes standards performance metrics such as precision, recall, Dice score, and Intersection over Union (IoU). The results exhibit a remarkable Dice coefficient score of 98.3% and an IoU rate surpassing 96.8%, demonstrating the model's robustness in identifying COVID-19-infected lung regions. © 2024 IEEE.Item Partial Convolution U-Net for Inpainting Distorted Images(Institute of Electrical and Electronics Engineers Inc., 2024) Rashmi Adyapady, R.; Annappa, B.; Sagar, P.Image inpainting is a domain in which researchers have shown considerable interest, and when it comes to deep learning techniques, realistic problems become interesting and challenging. In image inpainting, a corrupted facial image with missing holes or significant holes can be restored and compared to the original image to see if it is real or fake. In addition to fixing the texture of the image and getting the image's high-level abstract properties, it may also recover semantic images such as human faces. In the field of image-inpainting models, the Attention model with features learned through semantic approaches and progressive networks has become particularly popular. The proposed model introduces (i) Attention blocks in each decoder layer of U-Net architecture and (ii) a hybrid loss function leveraging both Mean Square Error (MSE) and Mean Absolute Error (MAE). The proposed Attention-based U-Net showed remarkable performance with SSIM and PSNR by 0.1067 and 13.63, respectively, compared to the previous approaches. © 2024 IEEE.Item Cognitive Chromatic Image Synthesis Using UNET and GAN(Institute of Electrical and Electronics Engineers Inc., 2024) Choubey, D.; Patil, V.; Anand Kumar, M.In the last decade, there has been a lot of interest for image colorization over a wide range of applications, especially in the restoration of old or damaged images. Because there are a lot of options when it comes to assigning color information, this problem is ill-posed by nature and is quite difficult to solve. Researchers have handled this issue in a variety of imaginative ways. More recent developments in automated colorization are focused on images that are repetitive in nature or images that require extensive editing. For instance, in such settings, semantic maps can be used as additional input to offer better control over the generalization of the colorization task with the help of conditional Deep Convolutional Generative Adversarial Networks (DCGANs).Our solution combines the techniques to allow computers to produce vivid visuals in this way. Monochrome or black and white images most of the times differ from the colored images in terms of visual detail and image content and colorizing them by hand is a tedious and often an artistic task. © 2024 IEEE.Item Semantic Segmentation of Underwater Images with CNN Based Adaptive Thresholding(Springer Science and Business Media Deutschland GmbH, 2025) Anand, S.K.; Kumar, P.V.; Saji, R.; Gadagkar, A.V.; Chandavarkar, B.R.Semantic segmentation remains a key research field in modern day computer vision and has been used in a myriad of applications across various fields. It can be extremely beneficial in the study of underwater scenes. Various underwater applications, like unmanned explorations and autonomous underwater vehicles, require accurate object classification and detection to allow the probes to avoid malicious objects. However, the models which work well for terrestrial images rarely work just as well for underwater images. This is because underwater images suffer from high blue light intensity as well as other ill-effects such as poor lighting and contrast. Trying to improve the model to account for bad image quality is not a great method as the model may misidentify noise as an image characteristic. In this paper, a unique CNN-based approach for post-processing image thresholding is proposed, on top of 3 models used for the semantic segmentation itself–Segnet, U-Net, and Deeplabv3+. The models’ outputs are then subject to the CNN-based post-processing technique to binarize the outputs into masks, and provides improved segmentation results compared to the base models. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.