China
Africa
Explore chapters and articles related to this topic
Multisensor Registration for Earth Remotely Sensed Imagery
Published in Rick S. Blum, Zheng Liu, Multi-Sensor Image Fusion and Its Applications, 2018
Jacqueline Le Moigne, Roger D. Eastman
For many applications of image processing, such as medical imagery, robotics, visual inspection and remotely sensed data processing, digital image registration is one of the first major processing steps. For all of these applications, image registration is defined as the process which determines the most accurate match between two or more images acquired at the same or different times by different or identical sensors. Registration provides the “relative” orientation of two images (or one image and other sources, e.g., a map), with respect to each other, from which the absolute orientation into an absolute reference system can be derived. As an illustration of this definition, Figure 11.1(a) shows an image extracted from a Landsat Thematic Mapper (TM) scene acquired over the Pacific Northwest. A corresponding image transformed by a rotation of 10° and a shift of 20 pixels horizontally and 60 pixels vertically is shown in Figure 11.1(b).
Scanning Techniques and Image Processing
Published in Yongjie Jessica Zhang, Geometric Modeling and Mesh Generation from Scanned Images, 2018
Image registration is the process of aligning or matching two images scanned at different time or using different modalities. Basically, it is to find the geometrical transform and map points from one image to another. Given a static or reference image S(x, y) and a moving or target image M(x,y), mathematically this can be written as () DetermineGandLsuchthatS(x,y)=G(M(L(x,y))),
Image Registration
Published in R. Suganya, S. Rajaram, A. Sheik Abdullah, Big Data in Medical Image Processing, 2018
R. Suganya, S. Rajaram, A. Sheik Abdullah
The primary goal of medical image registration is the process of geometrically aligning different sets of medical image data into one coordinate system. This alignment process requires the optimization of similarity measures. Mutual information (MI) is a popular entropy-based similarity measure which has found use in a large number of image registration applications. Stemming from information theory, this measure generally outperforms most other intensity-based measures in multimodal applications, as it does not assume the existence of any specific relationship between image intensities. It only assumes a statistical dependence. In this work, MI is selected in mono-modal applications to monitor the evolution of liver pathology diseases. The basic concept behind any approach using mutual information is to find a rigid body transformation, when applied to an image will maximize the MI between two images. While rigid body registration has become a widely used tool in clinical practice, non-rigid body registration has not yet achieved the same level for the surgeon during any decision making processes. Major difficulties in image registration process are image noise, illumination changes and occlusions. The examples for medical image registration is shown in Figure 22.
Multimodal MR image registration using weakly supervised constrained affine network
Published in Journal of Modern Optics, 2021
Xiaoyan Wang, Lizhao Mao, Xiaojie Huang, Ming Xia, Zheng Gu
Image registration methods based on deep learning can generally be divided into deformable and rigid registration. Deformable registration is used to find the association through an optimal nonlinear transformation, and is a fundamental method widely used in medical image analysis [3]. Fan et al. [4] proposed a BIRNet to perform brain image registration using a dual supervision learning strategy, they employed two ground truths generated by both ANTs [5] and LCC-Demons [6]. Unlike conventional methods to obtain ground truths described above, synthetic random transformations are used to train CNN. This method does not require manually annotated dataset and the output of the network is the displacement vector field on a thin plate spline transform grid [7,8]. The above-mentioned registration methods are all supervised. However, it is difficult to generate many matching clinical medical image pairs with known transformations.
Lung image registration by featured surface matching method
Published in Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 2022
Shu-Te Su, Ming-Chih Ho, Jia-Yush Yen, Yung-Yaw Chen
Deformable image registration can be applied to image-guided thoracoscopic surgery (Nakao et al. 2019). A preoperative organ image (surface plus internal information of the organ) is registered to an intraoperative organ image (partial surface or low-resolution 3D information) to produce the intraoperative underlying information. Intraoperative underlying information allows a surgeon to identify critical locations during surgery. It can be used for liver and lung resections to remove tumours and prevent injury to vessels. The technique is called Deformable Image Registration (DIR).s
An overview of deep learning methods for image registration with focus on feature-based approaches
Published in International Journal of Image and Data Fusion, 2020
Kavitha Kuppala, Sandhya Banda, Thirumala Rao Barige
The goal of image registration is to align two images of the same scene which differ due to geometric or photometric variations. Image registration is the first step in several application pipelines, such as change detection, image fusion across domains like remote sensing, stereo vision, medical imaging, etc. Classical image registration is broadly classified as area-based, feature-based and hybrid image registration (Zitová and Flusser 2003, Huang et al. 2004, Soh et al. 2014).