2017 14th IEEE India Council International Conference (INDICON), 2017
The two main tasks of computer vision problems are extraction of image descriptors and their subs... more The two main tasks of computer vision problems are extraction of image descriptors and their subsiquent matching. The choice of type and level of description is task dependent, which is a low-level approach for solving computer vision problems. Saliency detection is a basic task in perceptual organization. Without any prior assumption, automatic estimation of salient object regions in an image enhances the computer vision applications. In this paper, we proposed a novel saliency detection algorithm, which evaluates the local saliencies over random elliptical regions of interests. Though the proposed algorithm is multi-scale approach, it is simple, efficient, and gives high-quality saliency maps with higher resolution as compared to other existing methods. The qualitative and quantitative results of the proposed saliency deaction algorithm consistently outperforms other state-of-the atrs realizing higher precision with better recall rates and F-measure value.
For image fusion of geometrically distorted images, registration is the prerequisite step. Intens... more For image fusion of geometrically distorted images, registration is the prerequisite step. Intensity-based image registration methods are preferred due to higher accuracy than that of feature-based methods. But, perfect registered image using intensity based method leads towards improvements in computational complexity. Conventional transform like wavelet transform based image registration reduces the computational complexity, but suffers from discontinuities such as curved edges in the medical images. In this paper, a new registration algorithm is proposed that uses the approximate-level coefficients of the ripplet transform, which allows arbitrary support and degree as compared to curvelet transform. The entropy-based objective function is developed for registration using ripplet coefficients of the images. The computations are carried out with 6 sets of CT and MRI brain images to validate the performance of the proposed registration technique. The quantitative approach such as st...
Advances in Intelligent Systems and Computing, 2018
In intensity-based image registration methods, similarity measure plays a vital role. Recently, m... more In intensity-based image registration methods, similarity measure plays a vital role. Recently, mutual information and the variations of MI have gained popularity for the registration of multimodal images. As multimodal images have contrast changes, it is difficult to map them properly. To overcome this issue, phase congruency of the images that gives the significant features of illumination changed images. Also, the soft tissues present in the brain images have same intensity value in different regions. Hence, another assumption is that different pixels have unique distribution present in different regions for their proper characterization. For this challenge, utility measure is incorporated into enhanced mutual information as a weighted information to the joint histogram of the images. In this paper, spatial information along with features of phase congruency is combined to enhance the registration accuracy with less computational complexity. The proposed technique is validated with 6 sets of simulated brain images with different sets of transformed parameters. Evaluation parameters show the improvement of the proposed technique as compared to the other existing state of the arts.
In this paper, the problem of medical image segmentation is addressed in an unsupervised framewor... more In this paper, the problem of medical image segmentation is addressed in an unsupervised framework. We propose a novel method considering the hidden Markov random field model (HMRF) to model the image class labels, which takes into account the mutual influences of neighboring sites formulated on the basis of fuzzy clustering principle. The model parameters, number of class labels and the image labels are assumed to be unknown. Here an attempt has been made to incorporate the benefits of HMRF model into the benefits of fuzzy clustering procedure. To combine the spatial coherency modeling capabilities of the HMRF model and the enhanced flexibility obtained by fuzzy c-means (FCM) algorithm, fuzzy clustering expectation maximization (FCEM) algorithm is proposed. The initial model parameters are assumed arbitrarily unlike existing methods. Both model parameters as well as class labels of medical images are estimated recursively using proposed algorithm until the model parameters converge...
In medical imaging application, nonrigid registration is an important step which requires transfo... more In medical imaging application, nonrigid registration is an important step which requires transformation to align the deformed floating image grid points spatially with reference image grids. The transformation includes interpolation which estimates the intensity values of floating image other than the grid points. For nonrigid registration, generally the transformation is modeled using B-spline basis function. As the acquired images are deformed and contains bias field, it is very time consuming to reform the grid at each level. Considering the registration process as an optimization problem, the grid points are adaptively chosen and updated iteratively with a new transformation in each step. In this paper, an adaptive knot placement algorithm is presented for P-spline interpolation method to achieve accurate alignment of local deformation with less computation time. The proposed algorithm is validated with 3 sets of simulated brain images with a known distortion and 2 real brain MR image data sets. Different performance measures such as MSE, RMS, NAE are used to evaluate the qualitative measure of the proposed scheme. Estimated transformation grid and evaluated performance measures show the improvement in the proposed algorithm as compared to the other existing state-of-arts.
2017 14th IEEE India Council International Conference (INDICON), 2017
The two main tasks of computer vision problems are extraction of image descriptors and their subs... more The two main tasks of computer vision problems are extraction of image descriptors and their subsiquent matching. The choice of type and level of description is task dependent, which is a low-level approach for solving computer vision problems. Saliency detection is a basic task in perceptual organization. Without any prior assumption, automatic estimation of salient object regions in an image enhances the computer vision applications. In this paper, we proposed a novel saliency detection algorithm, which evaluates the local saliencies over random elliptical regions of interests. Though the proposed algorithm is multi-scale approach, it is simple, efficient, and gives high-quality saliency maps with higher resolution as compared to other existing methods. The qualitative and quantitative results of the proposed saliency deaction algorithm consistently outperforms other state-of-the atrs realizing higher precision with better recall rates and F-measure value.
For image fusion of geometrically distorted images, registration is the prerequisite step. Intens... more For image fusion of geometrically distorted images, registration is the prerequisite step. Intensity-based image registration methods are preferred due to higher accuracy than that of feature-based methods. But, perfect registered image using intensity based method leads towards improvements in computational complexity. Conventional transform like wavelet transform based image registration reduces the computational complexity, but suffers from discontinuities such as curved edges in the medical images. In this paper, a new registration algorithm is proposed that uses the approximate-level coefficients of the ripplet transform, which allows arbitrary support and degree as compared to curvelet transform. The entropy-based objective function is developed for registration using ripplet coefficients of the images. The computations are carried out with 6 sets of CT and MRI brain images to validate the performance of the proposed registration technique. The quantitative approach such as st...
Advances in Intelligent Systems and Computing, 2018
In intensity-based image registration methods, similarity measure plays a vital role. Recently, m... more In intensity-based image registration methods, similarity measure plays a vital role. Recently, mutual information and the variations of MI have gained popularity for the registration of multimodal images. As multimodal images have contrast changes, it is difficult to map them properly. To overcome this issue, phase congruency of the images that gives the significant features of illumination changed images. Also, the soft tissues present in the brain images have same intensity value in different regions. Hence, another assumption is that different pixels have unique distribution present in different regions for their proper characterization. For this challenge, utility measure is incorporated into enhanced mutual information as a weighted information to the joint histogram of the images. In this paper, spatial information along with features of phase congruency is combined to enhance the registration accuracy with less computational complexity. The proposed technique is validated with 6 sets of simulated brain images with different sets of transformed parameters. Evaluation parameters show the improvement of the proposed technique as compared to the other existing state of the arts.
In this paper, the problem of medical image segmentation is addressed in an unsupervised framewor... more In this paper, the problem of medical image segmentation is addressed in an unsupervised framework. We propose a novel method considering the hidden Markov random field model (HMRF) to model the image class labels, which takes into account the mutual influences of neighboring sites formulated on the basis of fuzzy clustering principle. The model parameters, number of class labels and the image labels are assumed to be unknown. Here an attempt has been made to incorporate the benefits of HMRF model into the benefits of fuzzy clustering procedure. To combine the spatial coherency modeling capabilities of the HMRF model and the enhanced flexibility obtained by fuzzy c-means (FCM) algorithm, fuzzy clustering expectation maximization (FCEM) algorithm is proposed. The initial model parameters are assumed arbitrarily unlike existing methods. Both model parameters as well as class labels of medical images are estimated recursively using proposed algorithm until the model parameters converge...
In medical imaging application, nonrigid registration is an important step which requires transfo... more In medical imaging application, nonrigid registration is an important step which requires transformation to align the deformed floating image grid points spatially with reference image grids. The transformation includes interpolation which estimates the intensity values of floating image other than the grid points. For nonrigid registration, generally the transformation is modeled using B-spline basis function. As the acquired images are deformed and contains bias field, it is very time consuming to reform the grid at each level. Considering the registration process as an optimization problem, the grid points are adaptively chosen and updated iteratively with a new transformation in each step. In this paper, an adaptive knot placement algorithm is presented for P-spline interpolation method to achieve accurate alignment of local deformation with less computation time. The proposed algorithm is validated with 3 sets of simulated brain images with a known distortion and 2 real brain MR image data sets. Different performance measures such as MSE, RMS, NAE are used to evaluate the qualitative measure of the proposed scheme. Estimated transformation grid and evaluated performance measures show the improvement in the proposed algorithm as compared to the other existing state-of-arts.
Uploads
Papers by Smita Pradhan