M Tech Dissertations

Permanent URI for this collectionhttp://ir.daiict.ac.in/handle/123456789/3

Browse

Has files: YesSubject: search.filters.subject.Image Processing

Search Results

Now showing 1 - 8 of 8
  • Thumbnail Image
    ItemOpen Access
    Single Image Dehazing for Homogeneous and Non-Homogeneous Haze using Deep Neural Network
    (2021) Gajjar, Manan; Mandal, Srimanta
    Image dehazing is an "ill-posed problem" that has been extensively studied in recent years. Available methods use various constraints/priors, deep learning, or a combination of both to get plausible dehazing solutions. This thesis work reviews some recent advancements and benchmarking, mainly focusing on proposed solutions and their results on both homogeneous and non-homogeneous haze datasets. Intending to achieve haze removal for both types of haze, we propose a new deep learning architecture. This convolution neural network is based on the reformulated atmospheric scattering model(ASM) and haze density estimation model to extract features for both types of haze. Model is trained on perceptual loss. Results on both indoor, outdoor homogeneous haze image datasets demonstrate our superior performance compared with other top deep learning architectures in terms of SSIM and PSNR. while on non-homogeneous haze dataset, the proposed model performs inferiorly compared to the state of the art non-homogeneous haze targeted dehazing model but much better than other homogeneous haze targeted dehazing models.
  • Thumbnail Image
    ItemOpen Access
    Medical image security with cheater identification using secret sharing scheme
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2015) Krishnan, Arun; Das, Manik Lal
    The progress in the field of information and communication technology has brought about quick and efficient transfer of textual as well as multimedia data . However, the growth of technology has also provided new ways for unauthorized access and illegal modifications of data, thereby, affecting data security. Secure data transmission is a necessity, especially in medical and legal fields. Doctors need to be convinced about the legitimacy of the medical images as well as associated health records they receive, especially in networking applications such as telediagnosis, teleconsultation,telesurgery etc. Furthermore, medical images should not be discernible to malignant agents with evil intentions on patient’s health. This thesis proposes a (k,n) secret sharing scheme for secure transfer of medical images and related electronic patient records (EPR) to a team of doctors through public insecure channels. The scheme prevents unauthorized access and detects illegal tampering of transmitted images and records. The scheme also considers the presence of a deceiver among the group of participants and includes techniques to detect any deception from the participant doctors and uniquely identify the deceiver. Furthermore, the scheme includes procedures to prevent the unauthorized release of medical image by any of the participating doctors. Simulation results shows that the proposed scheme satisfies all the security features discussed above.
  • Thumbnail Image
    ItemOpen Access
    Learning based approach for image compression
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2015) Kumar, Dheeraj; Joshi, Manjunath V.
    Data compression is a process of storing same information with less data or space in computer memory. There are many image compression techniques that are available for storing images with less storage space. Minimizing storage space minimizes the bandwidth required for transmission. In the proposed algorithm first level Discrete Wavelet Transformation (DWT, with Daubechies wavelets db4 as a mother wavelet) is applied on original image after which only low resolution coefficients are retained. Further Embedded Zero Tree Wavelet based algorithm (EZW) [10] is applied for best image quality at the given bit rate. We are using a set of images as database. For every input image Content Based Image Retrieval (CBIR) [7] technique is applied on database which results in some images, having similar content. At the receiver a learning based approach is used to decompress from resulted database images. Structure Similarity Index Measurement (SSIM) [15] an image quality assessment is used for similarity check. Inverse DWT is applied to get the estimate of the original. This is a lossy compression and results are compared with JPEG [13] and JPEG2000 [8] compression.
  • Thumbnail Image
    ItemOpen Access
    Study of fuzzy clustering algorithms and enhanced fuzzy reasoning application to texture based image segmentation
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2015) Gupta, Dhruv; Banerjee, Asim; Raval, Mehul S.; Shah, Pratik
    c-means (k-means) is a popular algorithm for cluster analysis. Many variants of c-means algorithms are available. All these models are studied in depth and convergence of iterative solutions are verified, in this thesis. An example of texture based image segmentation is used to support this study of various clustering algorithms. In context of clustering points in a space, a cluster represents a set of elements. The set is created by studying the membership of each element within it. Conventionally there are two types of set theories: crisp and its extension fuzzy set theory. The extension of crisp sets to fuzzy sets in terms of membership functions, is alike to extension of the set of integers to the set of real numbers. But the development does not end here, the membership can be extended to a vector value. Clustering is significantly affected by the data dimensionality and the distance metric used during cluster formation. Distance between points and distance between clusters are the key attributes for an accurate cluster analysis. During analysis of fuzzy based clustering a need for a new distance metric was felt. This metric defines distance between fuzzy sets and also between elements and fuzzy sets. As a step to fulfil this requirement, in this work the fuzzy sets with vector memberships are defined and proposed. Basic set theoretic operations, such as complement, union and intersection are defined and discussed in axiomatic manner. This work also proposes a new distance function defined for points and sets, and the new function is proved to be a metric through systematic proofs.
  • Thumbnail Image
    ItemOpen Access
    Image epitome generation and its applications
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2014) Shah, Twinkle; Mitra, Suman K.
    An image modeling scheme named “Epitome" was proposed by Brendan J. Frey and Nebjsa Jojic in 2002 [1]. Image epitome is a miniature, condensed version of the image. It is much smaller in the size compared to the input image but it contains the most constituent elements representing the image. Epitome has been used in various Image processing applications. It has been used for the Image processing applications like Image compression, Image denoising, Parts-based image retrieval, Image segmentation and Image in-painting. The estimation process of the epitome needs to be studied. The method proposed in [1] uses EM algorithm for epitome estimation. Instead of EM, some other estimation techniques can be applied to Epitome Generation. This modification to Epitome generation scheme may build better epitomes for different applications of Image epitome. It may happen that one epitome generation method generates epitomes which are more suitable for one application while less suitable for another application. In this thesis, a modification to the method proposed in [1] for Image denoising using epitome is proposed. This method gives a significant amount of improvement in the Image denoising. Also, two different parameter estimation techniques named DAEM (Deterministic Annealing EM) and Bayesian Sampling-Resampling approach are studied and applied for epitome generation. These methods are experimented for Image Reconstruction and Image denoising. DAEM gives improvement in Image denoising and Bayesian Learning based approach gives significant improvement in Image Reconstruction.
  • Thumbnail Image
    ItemOpen Access
    Text description of image
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2013) Sisodiya, Neha; Joshi, Manjunath V.
    Image comprises of something which is easy for human beings to understand but difficult for a machine to interpret. In this thesis, we propose an algorithm to obtain the textual description of the image content. In order to generate the output for a given image in terms of meaningful sentences that describes the image in the input, we have developed a stepwise procedure to fulfil the task. Problem statement is, given an image as an input, our system automatically generates the text description of the input image as output. Our aim is to understand scenario in an image i.e., describing given image automatically into simple sentences (English language). To accomplish our task four steps are involved 1) Segmentation 2) Recognition 3) Labelling 4) Sentence generation In first step segmentation is carried out using a novel approach of active contour model to separate the objects and background in the image. In order to separate the objects boundaries to get different regions present in the image first the segmentation is done which is helpful in the second step i.e., object Recognition. The object recognition is task of detecting and identifying objects in the scene depending on the feature vectors extracted from the image regions. We have extracted the features using SIFT (Scale Invariant Feature Transform) due to their invariant properties for recognition of an object. SIFT provides key point descriptors which we have used for labelling the object. In our method we try to recognize occluded and cluttered objects in the image and simultaneously improve segmentation by recognition and vice-a-versa. The next step is labelling the recognized objects i.e., which category the object belongs to and associate a label with it which is useful in next step i.e., generation of sentences. We have used SVM (Support Vector Machine) classifier for classifying the objects. Our final step involves generation and this is accomplished by linking labels by their meanings and form meaningful sentences as an output of our system.
  • Thumbnail Image
    ItemOpen Access
    Texton based auto region detection for image inpainting
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2012) Vora, Manali; Joshi, Manjunath V.
    Historical monuments are considered as one of the key aspects for modern communities. Unfortunately, because of variety of factors these monuments are sometimes damaged or destroyed. Image inpainting is the process of restoring the damaged image and hence can be used as a useful tool for restoring the images of historical monumnets. Inpainting techniques developed so far require the user to manually select regions to be inpainted. In this thesis, we propose a novel approach for automatic region detection for inpainting. Given a frontal face test image and a set of face images of monument consisting of vandalized and nonvandalized regions, our task is to: 1. extract potential regions of interest like eyes, nose and lips, 2. identify whether a particular region is vandalized or not and 3. inpaint the vandalized regions using the available non-vandalized regions. In our approach, potential regions of interest are rst localized using the bilateral symmetry based method, while the identication of vandalized and non-vandalized regions is done based on the texture statistics. The texture statistics are obtained by extracting the textons from the lter response space by using the Kmeans algorithm. After identifying vandalized regions, Poisson image editing technique is used to inpaint them using the non-vandalized regions available either in the same image or from the other images in the database. Novelty of our approach lies in 1. automatic detection of target regions for inpaiting and 2. automatic selection of optimum number of textons. Experiments conducted on the frontal face images of monuments downloaded from the Internet give promising results
  • Thumbnail Image
    ItemOpen Access
    Automatic image inpainting for the facial images of monuments
    (Dhirubhai Ambani Institute of Information and Communication Technology, 2011) Parmar, Chintan; Joshi, Manjunath V.
    Image inpainting is the process of restoring or modifying the image contents im- perceptibly. In this paper we propose a novel approach for automatic image inpainting of the ruined frontal face images of monuments or statues of historic places. Given a facial test image and a set of frontal face images consisting of vandalized and non-vandalized regions captured from the same historical site, our objective is to: 1. detect both the vandalized and non-vandalized regions, 2. inpaint the vandalized regions using the non-vandalized one. Potential regions of interest (ROI) are localized using edge based features. Final detection is done using appropriate designed templates for vandalized and non-vandalized regions. Once the regions are detected the inpainting of the vandalized region is done by using similar regions that are non-vandalized. The non-vandalized regions are selected either from the same image or from other images available in the database. Poisson image editing is used for inpainting the vandalized regions. Novelty of our approach lies in automatic detection of vandalized regions and inpainting them using the non-vandalized regions already existing in the database.