Faculty Publications
Permanent URI for this communityhttps://idr.nitk.ac.in/handle/123456789/18736
Publications by NITK Faculty
Browse
64 results
Search Results
Item An approach for mining heterogeneous data for cross-media retrieval(2013) Pavan, K.M.; Ananthanarayana, V.S.Due to the wide availability of huge amount of multimedia data in various modalities such as image and text documents, having a great amount of similarity among them is inevitable. In this paper, we present an efficient model which correlates the similarity among documents belonging to various modalities to achieve cross-media retrieval. Cross-media retrieval is a content based information retrieval system where heterogeneous data is mined to retrieve results of various modalities, i.e., input object and returned results may be of different modalities. For example, text objects can be retrieved as a result to image input. First, features are extracted from multimedia objects by which the objects are labeled. Using the labels, similar documents are grouped to generate Multimedia Documents. We construct a cross-media correlation graph with documents as vertices, where positive weight is assigned to every single edge according to the amount of similarity between vertices. The cross-media retrieval system identifies the input document and as a result returns required number of documents with highest weights. © 2013 IEEE.Item Semi-supervised Semantic Segmentation of Effusion Cytology Images Using Adversarial Training(Springer Science and Business Media Deutschland GmbH, 2023) Rajpurohit, M.; Aboobacker, S.; Vijayasenan, D.; Sumam David, S.; Suresh, P.K.; Sreeram, S.In pleural effusion, an excessive amount of fluid gets accumulated inside the pleural cavity along with signs of inflammation, infections, malignancies, etc. Usually, a manual cytological test is performed to detect and diagnose pleural effusion. The deep learning solutions for effusion cytology include a fully supervised model trained on effusion cytology images with the help of output maps. The low-resolution cytology images are harder to label and require the supervision of an expert, the labeling process time-consuming and expensive. Therefore, we have tried to use some portion of data without any labels for training our models using the proposed semi-supervised training methodology. In this paper, we proposed an adversarial network-based semi-supervised image segmentation approach to automate effusion cytology. The semi-supervised methodology with U-Net as the generator shows nearly 12% of absolute improvement in the f-score of benign class, 8% improvement in the f-score of malignant class, and 5% improvement in mIoU score as compared to a fully supervised U-Net model. With ResUNet++ as a generator, a similar improvement in the f-score of 1% for benign class, 8% for the malignant class, and 1% in the mIoU score is observed as compared to a fully supervised ResUNet++ model. © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.Item Face Recognition (FR) systems are increasingly gaining more importance. Face detection and tracking in a complex scene forms the first step in building a practical FR system. In this paper, a method to detect and track human faces in color image sequences is described. Skin color classification and morphological segmentation is used to detect face(s) in the first frame. These detected faces are tracked over subsequent frames by using the position of the faces in the first frame as the marker and detecting for skin in the localized region. Specific advantages of this approach are that skin color analysis method is simple and powerful, and the system can be used to detect/track multiple faces. © 2002 Taylor & Francis Group, LLC.(Human face detection and tracking using skin color modeling and connected component operators) Kuchi, P.; Gabbur, P.; Subbanna Bhat, P.; Sumam David, S.2002Item In this paper, block based texture segmentation is proposed based on contourlets and the hidden Markov model (HMM). Hidden Markov model is combined with hidden Markov tree (HMT) to form HMM-HMT model that models global dependency between the blocks in addition to the local statistics within a block. The HMM-HMT model is modified to use the contourlet transform, a new extension to the wavelet transform that forms a true basis for image representations. The maximum likelihood multiresolution segmentation algorithm is used to handle several block sizes at once. Since the algorithm works on the contourlet transformed image data, it can directly segment images without the need for transforming into the space domain. The experimental results demonstrate the competitive performance of the algorithm on contourlets with that of the other methods and excellent visual performance at small block sizes. The performance is comparable with that of wavelets and is superior at small block sizes. © Springer-Verlag 2004.(Springer Verlag, Contourlet based multiresolution texture segmentation using contextual hidden markov models) Raghavendra, B.S.; Subbanna Bhat, P.2004Item Content-based image retrieval in presence of foreground disturbances(2006) Shekhar, R.; Chaudhuri, S.The image retrieval problem in the presence of possible foreground disturbances is considered. The foreground may be irrelevant for the retrieval purposes but it occludes the background and hence reduces the retrieval accuracy. The accrued inaccuracy is quantified in terms of cardinality of the occluding region. The use of the sprite generated from a video clip is proposed as a query, so that the effect of moving foreground can be eliminated. The segmented foreground region is filled in by constructing the background sprite to increase the retrieval accuracy. The performance of a retrieval scheme under foreground disturbance is presented here. © The Institution of Engineering and Technology 2006.Item Despeckling low SNR, low contrast ultrasound images via anisotropic level set diffusion(Kluwer Academic Publishers, 2014) Bini, A.A.; Bhat, M.S.Speckle is a form of multiplicative and locally correlated noise which degrades the signal-to-noise ratio (SNR) and contrast resolution of ultrasound images. This paper presents a new anisotropic level set method for despeckling low SNR, low contrast ultrasound images. The coefficient of variation, a speckle-robust edge detector is embedded in the well known geodesic "snakes" model to smooth the image level sets, while preserving and sharpening edges of a speckled image. The method achieves much better speckle suppression and edge preservation compared to the traditional anisotropic diffusion based despeckling filters. In addition, the performance of the filter is less sensitive to the speckle scale of the image and edge contrast parameter, which makes it more suitable for the detection of low contrast features in an ultrasound image. We validate the method using both synthetic and real ultrasound images and quantify the performance improvement over other state-of-the-art algorithms in terms of speckle noise reduction and edge preservation indices. © 2012 Springer Science+Business Media, LLC.Item Depth-Based Selective Blurring in Stereo Images Using Accelerated Framework(3D Research Center 3drc@kw.ac.kr, 2014) Mukherjee, S.; Guddeti, R.M.R.Abstract: We propose a hybrid method for stereo disparity estimation by combining block and region-based stereo matching approaches. It generates dense depth maps from disparity measurements of only 18 % image pixels (left or right). The methodology involves segmenting pixel lightness values using fast K-Means implementation, refining segment boundaries using morphological filtering and connected components analysis; then determining boundaries’ disparities using sum of absolute differences (SAD) cost function. Complete disparity maps are reconstructed from boundaries’ disparities. We consider an application of our method for depth-based selective blurring of non-interest regions of stereo images, using Gaussian blur to de-focus users’ non-interest regions. Experiments on Middlebury dataset demonstrate that our method outperforms traditional disparity estimation approaches using SAD and normalized cross correlation by up to 33.6 % and some recent methods by up to 6.1 %. Further, our method is highly parallelizable using CPU–GPU framework based on Java Thread Pool and APARAPI with speed-up of 5.8 for 250 stereo video frames (4,096 × 2,304). © 2014, 3D Research Center, Kwangwoon University and Springer-Verlag Berlin Heidelberg.Item Self-optimal clustering technique using optimized threshold function(Institute of Electrical and Electronics Engineers Inc., 2014) Verma, N.K.; Roy, A.This paper presents a self-optimal clustering (SOC) technique which is an advanced version of improved mountain clustering (IMC) technique. The proposed clustering technique is equipped with major changes and modifications in its previous versions of algorithm. SOC is compared with some of the widely used clustering techniques such as K-means, fuzzy C-means, Expectation and Maximization, and K-medoid. Also, the comparison of the proposed technique is shown with IMC and its last updated version. The quantitative and qualitative performances of all these well-known clustering techniques are presented and compared with the aid of case studies and examples on various benchmarked validation indices. SOC has been evaluated via cluster compactness within itself and separation with other clusters. The optimizing factor in the threshold function is computed via interpolation and found to be effective in forming better quality clusters as verified by visual assessment and various standard validation indices like the global silhouette index, partition index, separation index, and Dunn index. © 2007-2012 IEEE.Item Computer-aided diagnosis system for tissue characterization of brain tumor on magnetic resonance images(Springer London, 2015) Arakeri, M.P.; Guddeti, G.R.M.The manual analysis of brain tumor on magnetic resonance (MR) images is time-consuming and subjective. Thus, to avoid human errors in brain tumor diagnosis, this paper presents an automatic and accurate computer-aided diagnosis (CAD) system based on ensemble classifier for the characterization of brain tumors on MR images as benign or malignant. Brain tumor tissue was automatically extracted from MR images by the proposed segmentation technique. A tumor is represented by extracting its texture, shape, and boundary features. The most significant features are selected by using information gain-based feature ranking and independent component analysis techniques. Next, these features are used to train the ensemble classifier consisting of support vector machine, artificial neural network, and k-nearest neighbor classifiers to characterize the tumor. Experiments were carried out on a dataset consisting of T1-weighted post-contrast and T2-weighted MR images of 550 patients. The developed CAD system was tested using the leave-one-out method. The experimental results showed that the proposed segmentation technique achieves good agreement with the gold standard and the ensemble classifier is highly effective in the diagnosis of brain tumor with an accuracy of 99.09 % (sensitivity 100 % and specificity 98.21 %). Thus, the proposed system can assist radiologists in an accurate diagnosis of brain tumors. © 2013, Springer-Verlag London.Item An efficient framework for segmentation and identification of tumours in brain MR images(Inderscience Enterprises Ltd., 2016) Parameshwari, D.S.; Aparna., P.In this research work, two efficient textural feature extraction (TFE) algorithms (TFEA-I and TFEA-II) are proposed for a class of brain magnetic resonance imaging (MRI) applications. TFEA-I employs higher order statistical cumulant, namely, Kurtosis in order to generate a feature set based on the probability density function (PDF) of generalised Gaussian model that represents thewavelet coefficient energies of the sub-bands of decomposed image. TFEA-II derives a feature set employing cooccurrence matrix model for second order statistical characterisation of wavelet coefficients. In conjunction with TFEA-I and TFEA-II, we propose segmentation framework to compute coarse and smooth segmented boundaries for the tumour. When compared with the conventional TFEA methods reported in the literature, the use of proposed TFEA-I and TFEA-II results in two important advantages; considerable reduction in the feature set size and elimination of the need for using specialised feature selection/reduction algorithms thereby making them highly attractive for a class of brain MR imaging application. © © 2016 Inderscience Enterprises Ltd.