Figure 5 - uploaded by Lewis D Griffin
Content may be subject to copyright.
Source publication
We propose a new voxel similarity measure which uses local image structure as well as intensity information. The derivatives of linear scale space are used to provide struc-tural information in the form of a feature vector for each voxel. Each scale space derivative is assigned to its own in-formation channel. We illustrate the behavior of the simi...
Contexts in source publication
Context 1
... images can be considered as a registration gold-standard, and the graphs of the registra- tion function tell us how the similarity measure behaves as a function of misregistration for images with a noise differ- ence. We took an axial slice through the lateral ventricles and extracted a 32 × 32 pixel sub-image as illustrated in Figure 5. Then we misregistered the sub-image relative to the other image by applying a x-translation t x , t x increases from left to right direction in Figure 5. t x = 0 voxels rep- resents perfect registration. ...
Context 2
... images can be considered as a registration gold-standard, and the graphs of the registra- tion function tell us how the similarity measure behaves as a function of misregistration for images with a noise differ- ence. We took an axial slice through the lateral ventricles and extracted a 32 × 32 pixel sub-image as illustrated in Figure 5. Then we misregistered the sub-image relative to the other image by applying a x-translation t x , t x increases from left to right direction in Figure 5. t x = 0 voxels rep- resents perfect registration. ...
Similar publications
We propose a workflow which includes the essential step of feature selection in order to optimize combined density functional theory—machine learning schemes (DFT-ML). Here, the energy gaps of hybrid graphene—boron nitride nanoflakes with randomly distributed domains are predicted using artificial neural networks (ANNs). The training data is obtain...
With the bag-of-contextual-visual-word (BOCVW) models, we propose a scene categorization method based on local–global feature fusion and multi-scale multi-spatial resolution encoding. First, the performances of the BOCVW models belonging to different features are mutually reinforced by fusing other types of features within local regions. Then, the...
In the complex electromagnetic environment, the large number of pulse loss (data loss) and the serious overlap of signal parameter space destroy the signal regularity used by the traditional sorting method and finally make the existing signal sorting algorithm difficult to obtain satisfactory sorting effect or even completely invalid. On the basis...
In order to enhance the surveillance system, detecting human and acceding their behavior is a must requirement. This research paper is presenting a methodology towards improvement of human detection approaches. Human dataset is created through 2D images by developing feature vector of humans using various image processing techniques in order to ext...
Citations
... The powerful fitting ability of neural networks enables things that traditional machine learning cannot accomplish. Recent works have introduced mutual information theory to image and also achieved excellent results in vision tasks, e.g., feature classification and feature selection in [29][30][31][32], medical image registration in [33,34], image classification in [35,36] and image similarity metrics in [37]. ...
A large number of high-quality face swapping images are often required to improve the performance of forgery detection. High-quality face swapping images require fewer face swapping traces and fewer face swapping artificialities in the image while ensuring the same identity consistency with the source face and the same attribute consistency with the target. It is a challenging task today to properly separate identity and non-identity related attribute information. In this paper, we propose a novel framework that called high-fidelity face swapping algorithm (HFSA) which consists of two part networks, a GAN-based mutual information swapping network, MuIn-swap, for face swapping and an MAE-based Detail Repair Net, DRN, for detail repair. We introduce mutual information into feature compression, explicitly computing the mutual information of identity and attribute information obtained by compression of latent features. Therefore, the learning of the network is explicitly guided by formulating the minimum mutual information as the optimization goal, so that we can obtain pure identity and attribute information. In addition to overcome the problem of information loss during face swapping, we additionally design an DRN to repair the details of the face swapping to achieve more realistic and fidelity face swapping images. Through extensive experiments, it has been demonstrated that the forgery samples generated by HFSA guarantee smaller forgery traces and fewer artifacts while guaranteeing identity consistency with the source and attribute preservation. The code is already available on GitHub: https://github.com/karasuma123/HFSA.
... It is based on generating a one parameter family of images by blurring the image with Gaussian kernels of increasing width (the scale parameter) and analyzing these blurred images to extract structural features [7]. Scale space images at one value of scale parameter and their derivatives were used as features in [8] for MI based registration. In this work, we define the features for MI based priors as the original image, images blurred at different scales, and the Laplacians of the blurred images. ...
We propose a mutual information based prior for incorporating information from co-registered anatomical images into PET image reconstruction. The prior uses mutual information between feature vectors that are extracted from the anatomical and functional images using a scale space approach. We perform simulations on a realistic 3D phantom generated by replicating a 2-D autoradiographic cross section of a mouse labelled with F18-FDG. A digital photograph of the cryosection of the same slice is used to generate the anatomical image. The images are registered using mutual information based rigid registration. PET data are then simulated from the autoradiography based phantom. We use a preconditioned conjugate gradient algorithm to compute the PET image that maximizes the posterior density. The performance of this method is compared with that using a Gaussian quadratic penalty, which does not use anatomical information. Simulation results indicate that the mutual information based prior can achieve reduced standard deviation at comparable bias compared to the quadratic penalty
... Another limitation of these methods is that the computation of the joint histogram is calculated from the correspondence between individual voxels in the two images. In recent years, different approaches have considered a region-based correlation to compute image similarity [6, 7, 8, 9, 10]. In this paper, we propose a new MI-based framework that uses structural information in an image. ...
... They assume that the high-dimensional distribution is approximately normally distributed . Holden et al. [10] use the derivatives of gaussian scale space to provide structural information in the form of a feature vector for each voxel. ...
Mutual information has been successfully used as an effective similarity measure for multimodal image registration. However,
a drawback of the standard mutual information-based computation is that the joint histogram is only calculated from the correspondence
between individual voxels in the two images. In this paper, the normalized mutual information measure is extended to consider
the correspondence between voxel blocks in multimodal rigid registration. The ambiguity and high-dimensionality that appears
when dealing with the voxel neighborhood is solved using uniformly distributed random lines and reducing the number of bins
of the images. Experimental results show a significant improvement with respect to the standard normalized mutual information.
... It is known that higher-order image signal statistics exist and can be used to describe a voxel [58], [59], [81][82][83][84][85]. Similarity measures that compare the distribution of signal intensities within a local neighbourhood about a voxel of interest are more robust to the uncertainties discussed previously [86][87][88][89][90][91]. This robustness is the result of using a normalized distribution of signal intensities in the local neighbourhood about the voxel of interest to characterize local image structure. ...
... This can be achieved by employing a multi-resolution architecture, which has been used to encode spatial information in many intensity-based similarity measures [86], [91], [92], [95]. If histogram moments in an image sub-region centered on a voxel of interest are computed over multiple image resolutions, the histograms will capture information about that voxel's global and local structural context. ...
Our intention was to register combined FDG PET-CT head and neck images from multiple patients to a template image. We used a free-form deformation (FFD) nonrigid registration (NRR) algorithm. We aimed to overcome instances of mis-registration at the coarse scale that occur when using a similarity measure derived from the CT images alone. We proposed a multi-component (MC) similarity measure to improve the robustness of the registration by utilizing information from both the PET and the CT image. Information from all combinations of the dual modality dataset were incorporated into this measure. We defined the MC similarity measure to be a weighted sum of four components: the normalized sum of squared differences (SSD) for the CT-CT component and normalized mutual information (NMI) for the other components. This function was then used to register a group of combined PET-CT images demonstrating normal anatomy and FDG uptake to the template image. The performance of the registration was assessed by measuring the errors across three bony anatomical landmarks within the image. The results were compared to the registration using SSD of the CT images alone as a similarity measure. Registrations performed using this MC similarity measure demonstrated improved robustness compared to using the SSD of the CT images alone as a similarity measure.
Information .eory (IT) tools, widely used in many scientific fields such as engineering, physics, genetics, neuroscience, and many others, are also useful transversal tools in image processing. In this book, we present the basic concepts of IT and how they have been used in the image processing areas of registration, segmentation, video processing, and computational aesthetics. Some of the approaches presented, such as the application of mutual information to registration, are the state of the art in the field. All techniques presented in this book have been previously published in peer-reviewed conference proceedings or international journals. We have stressed here their common aspects, and presented them in an unified way, so to make clear to the reader which problems IT tools can help to solve, which specific tools to use, and how to apply them. .e IT basics are presented so as to be self-contained in the book. .e intended audiences are students and practitioners of image processing and related areas such as computer graphics and visualization. In addition, students and practitioners of IT will be interested in knowing about these applications.
Image registration consists in finding the transformation that brings one image into the best possible spatial correspondence with another image. In this paper, we present a new framework for image registration based on compression. The basic idea underlying our approach is the conjecture that two images are correctly registered when we can maximally compress one image given the information in the other. The contribution of this paper is twofold. First, we show that image registration can be formulated as a compression problem. Second, we demonstrate the good performance of the similarity metric, introduced by Li et al., in image registration. Two different approaches for the computation of this similarity metric are described: the Kolmogorov version, computed using standard real-world compressors, and the Shannon version, calculated from an estimation of the entropy rate of the images.
Purpose:
Tumor characterization employing a voxel-wise analysis of image signal facilitates the determination of the spatial distribution of tumor attributes, and when employed for therapy response assessment offers the promise of greater sensitivity to change than conventional approaches. However, the accuracy of a voxel-wise analysis of change is limited by local registration uncertainties that can disrupt the spatiotemporal correspondence between assessed voxels. We present a method for assessing voxel correspondence strength using a multiresolution local histogram-based measure of image structure similarity. When employed in a longitudinal tumorimaging context, a voxel similarity measure must be robust to intensity variations that can arise from the image acquisition, treatment effects, or changes in underlying disease processes. Consequently, the local histogram-based similarity measure is evaluated for sensitivity to structural change and robustness to intensity variation and is compared against normalized mutual information and normalized cross-correlation.
