• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.1
• /
• pp.180-194
• /
• 2021

For small image registration, feature-based approaches are likely to fail as feature detectors cannot detect enough feature points from low-resolution images. The classic FFT approach's prediction accuracy is high, but the registration time can be relatively long, about several seconds to register one image pair. To achieve real-time and high-precision rigid registration for small images, we apply deep neural networks for supervised rigid transformation prediction, which directly predicts the transformation parameters. We train deep registration models with rigidly transformed CIFAR-10 images and STL-10 images, and evaluate the generalization ability of deep registration models with transformed CIFAR-10 images, STL-10 images, and randomly generated images. Experimental results show that the deep registration models we propose can achieve comparable accuracy to the classic FFT approach for small CIFAR-10 images (32×32) and our LSTM registration model takes less than 1ms to register one pair of images. For moderate size STL-10 images (96×96), FFT significantly outperforms deep registration models in terms of accuracy but is also considerably slower. Our results suggest that deep registration models have competitive advantages over conventional approaches, at least for small images.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.6
• /
• pp.652-658
• /
• 2019

When IoT is applied to various research experiment fields such as physics, pharmacy, biology and medicine, it can increase the safety and convenience of researchers by intelligently monitoring and controlling research equipment and environment with various sensors and devices. For accurate and convenient record management and the research history and the basis but also for the reverse tracking, real-time reagent measurement and registration should be provided as a research support automation services. Currently, existing methods of reagent management are operated by computerized method, but reagent registration and management are not automated. And also record is managed manually, there are many hassles and problems such as a record error and too much time required for quantification and registration for many reagents. In this paper, we study a real time reagent measuring and registration method based on IoT to resolve the problems aforementioned, by the information of the reagent acquired by image recognition and NFC method.

• Journal of Radiation Protection and Research
• /
• v.44 no.4
• /
• pp.149-155
• /
• 2019

Background: Magnetic resonance (MR) image guided radiation therapy system, enables real time MR guided radiotherapy (RT) without additional radiation exposure to patients during treatment. However, MR image lacks electron density information required for dose calculation. Image fusion algorithm with deformable registration between MR and computed tomography (CT) was developed to solve this issue. However, delivered dose may be different due to volumetric changes during image registration process. In this respect, synthetic CT generated from the MR image would provide more accurate information required for the real time RT. Materials and Methods: We analyzed 1,209 MR images from 16 patients who underwent MR guided RT. Structures were divided into five tissue types, air, lung, fat, soft tissue and bone, according to the Hounsfield unit of deformed CT. Using the deep learning model (U-NET model), synthetic CT images were generated from the MR images acquired during RT. This synthetic CT images were compared to deformed CT generated using the deformable registration. Pixel-to-pixel match was conducted to compare the synthetic and deformed CT images. Results and Discussion: In two test image sets, average pixel match rate per section was more than 70% (67.9 to 80.3% and 60.1 to 79%; synthetic CT pixel/deformed planning CT pixel) and the average pixel match rate in the entire patient image set was 69.8%. Conclusion: The synthetic CT generated from the MR images were comparable to deformed CT, suggesting possible use for real time RT. Deep learning model may further improve match rate of synthetic CT with larger MR imaging data.

• Journal of KIISE:Software and Applications
• /
• v.33 no.3
• /
• pp.335-343
• /
• 2006

This paper proposes a novel technique of marker-based 2D-3D registration to combine 3D information obtained from preoperative CT images into 2D image obtained from intraoperative x-ray fluoroscopy image. Our method is divided into preoperative and intraoperative procedures. In preoperative procedure, we generate CT-derived DRRs using graphics hardware and detect markers automatically. In intraoperative procedure, we propose a hierarchical two- step registration to reduce a degree of freedom from 6-DOP to 2-DOF which is composed of in-plane registration using principal axis method and out-plane registration using minimal error searching method in spherical coordinate. For experimentation, we use cardiac phantom datasets with confirmation markers and evaluate our method in the aspects of visual inspection, accuracy and processing time. As experimental results, our method keeps accuracy and aligns very fast by reducing real-time computations.

• Journal of Korea Multimedia Society
• /
• v.17 no.7
• /
• pp.818-828
• /
• 2014

After emerging of Microsoft Kinect, the interest in three-dimensional (3D) depth image was significantly increased. Depth image data of an object can be converted to 3D coordinates by simple arithmetic calculation and then can be reconstructed as a 3D model on computer. However, because the surface coordinates can be acquired only from the front area facing Kinect, total solid which has a closed surface cannot be reconstructed. In this paper, 3D registration method for multiple Kinects was suggested, in which surface information from each Kinect was simultaneously collected and registered in real time to build 3D total solid. To unify relative coordinate system used by each Kinect, 3D perspective transform was adopted. Also, to detect control points which are necessary to generate transformation matrix, 3D randomized Hough transform was used. Once transform matrices were generated, real time 3D reconstruction of various objects was possible. To verify the usefulness of suggested method, human arms were 3D reconstructed and the volumes of them were measured by using four Kinects. This volume measuring system was developed to monitor the level of lymphedema of patients after cancer treatment and the measurement difference with medical CT was lower than 5%, expected CT reconstruction error.

• Journal of Biomedical Engineering Research
• /
• v.25 no.2
• /
• pp.111-117
• /
• 2004

Most of digital subtraction methods in dental radiography are based on registration using manual landmarks. We have developed an automatic registration method without using the manual selection of landmarks. By restricting a geometrical matching of images to a region of interest (ROl), we compare the cross-correlation coefficient only between the ROIs. The affine or perspective transform parameters satisfying maximum of cross-correlation between the local regions are searched iteratively by a fast searching strategy. The parameters are searched on the 1/4 scale image coarsely and then, the fine registration is performed on the original scale image. The developed method can match the images corrupted by Gaussian noise with the same accuracy for the images without any transform simulation. The registration accuracy of the perspective method shows a 17% improvement over the manual method. The application of the developed method to radiography of dental implants provides an automatic noise robust registration with high accuracy in almost real time.

• Journal of the Institute of Electronics Engineers of Korea CI
• /
• v.45 no.3
• /
• pp.97-102
• /
• 2008

In this paper we propose an object-based image retrieval method using spatial color model and feature points registration method for an effective color query detection. The proposed method in other to overcome disadvantages of existing color histogram methods and then this method is use the HMMD model and rough set in order to segment and detect the wanted image parts as a real time without the user's manufacturing in the database image and query image. Here, we select candidate regions in the similarity between the query image and database image. And we use SIFT registration methods in the selected region for object retrieving. The experimental results show that the proposed method is more satisfactory detection radio than conventional method.

• Journal of Computing Science and Engineering
• /
• v.6 no.1
• /
• pp.1-11
• /
• 2012

Computed tomography (CT) images are widely used for the analysis of the temporal evaluation or monitoring of the progression of a disease. The follow-up examinations of CT scan images of the same patient require a 3D registration technique. In this paper, an automatic and robust registration is proposed for the rigid registration of 3D CT images. The proposed method involves two steps. Firstly, the two CT volumes are aligned based on their principal axes, and then, the alignment from the previous step is refined by the optimization of the similarity score of the image's voxel. Normalized cross correlation (NCC) is used as a similarity metric and a downhill simplex method is employed to find out the optimal score. The performance of the algorithm is evaluated on phantom images and knee synthetic CT images. By the extraction of the initial transformation parameters with principal axis of the binary volumes, the searching space to find out the parameters is reduced in the optimization step. Thus, the overall registration time is algorithmically decreased without the deterioration of the accuracy. The preliminary experimental results of the study demonstrate that the proposed method can be applied to rigid registration problems of real patient images.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.49 no.4
• /
• pp.10-16
• /
• 2012

For high speed visual inspection in semiconductor industries, it is essential to acquire two-dimensional images on regions of interests with a large field of view (FOV) and a high resolution simultaneously. In this paper, an imaging system is newly proposed to achieve high quality image in terms of precision and FOV, which is composed of single lens, a beam splitter, two camera sensors, and stereo image grabbing board. For simultaneously acquired object images from two camera sensors, Zhang's camera calibration method is applied to calibrate each camera first of all. Secondly, to find a mathematical mapping function between two images acquired from different view cameras, the matching matrix from multiview camera geometry is calculated based on their image homography. Through the image homography, two images are finally registered to secure a large inspection FOV. Here the inspection system of using multiple images from multiple cameras need very fast processing unit for real-time image matching. For this purpose, parallel processing hardware and software are utilized, such as Compute Unified Device Architecture (CUDA). As a result, we can obtain a matched image from two separated images in real-time. Finally, the acquired homography is evaluated in term of accuracy through a series of experiments, and the obtained results shows the effectiveness of the proposed system and method.

• Journal of KIISE
• /
• v.42 no.8
• /
• pp.990-997
• /
• 2015

Panoramic image creation has been extensively studied. Existing methods use customized hardware, or apply post-processing methods to seamlessly stitch images. These result in an increase in either cost or complexity. In addition, images can only be stitched under certain conditions such as existence of characteristic points of the images. This paper proposes a low cost and easy-to-use system that produces realtime panoramic video. We use an off-the-shelf embedded platform to capture multiple images, and these are then transmitted to a server in a compressed format to be merged into a single panoramic video. Finally, we analyze the performance of the implemented system by measuring time to successfully create the panoramic image.