• Journal of Biomedical Engineering Research
• /
• v.44 no.5
• /
• pp.354-361
• /
• 2023

Most of spinal diseases are diagnosed based on the subjective judgment of a specialist, so numerous studies have been conducted to find objectivity by automating the diagnosis process using deep learning. In this paper, we propose a method that combines segmentation and feature extraction, which are frequently used techniques for diagnosing spinal diseases. Four models, U-Net, U-Net++, DeepLabv3+, and M-Net were trained and compared using 1000 X-ray images, and key-points were derived using Douglas-Peucker algorithms. For evaluation, Dice Similarity Coefficient(DSC), Intersection over Union(IoU), precision, recall, and area under precision-recall curve evaluation metrics were used and U-Net++ showed the best performance in all metrics with an average DSC of 0.9724. For the average Euclidean distance between estimated key-points and ground truth, U-Net was the best, followed by U-Net++. However the difference in average distance was about 0.1 pixels, which is not significant. The results suggest that it is possible to extract key-points based on segmentation and that it can be used to accurately diagnose various spinal diseases, including spondylolisthesis, with consistent criteria.

• International Journal of Internet, Broadcasting and Communication
• /
• v.12 no.4
• /
• pp.83-92
• /
• 2020

A recent study to determine the fall is focused on analyzing fall motions using a recurrent neural network (RNN), and uses a deep learning approach to get good results for detecting human poses in 2D from a mono color image. In this paper, we investigated the improved detection method to estimate the position of the head and shoulder key points and the acceleration of position change using the skeletal key points information extracted using PoseNet from the image obtained from the 2D RGB low-cost camera, and to increase the accuracy of the fall judgment. In particular, we propose a fall detection method based on the characteristics of post-fall posture in the fall motion analysis method and on the velocity of human body skeleton key points change as well as the ratio change of body bounding box's width and height. The public data set was used to extract human skeletal features and to train deep learning, GRU, and as a result of an experiment to find a feature extraction method that can achieve high classification accuracy, the proposed method showed a 99.8% success rate in detecting falls more effectively than the conventional primitive skeletal data use method.

• Journal of the Korea Computer Graphics Society
• /
• v.29 no.1
• /
• pp.1-11
• /
• 2023

Deep-learning-based image segmentation is one of the most widely employed lane detection approaches, and it requires a post-process for extracting the key points on the lanes. A general approach for key-point extraction is using a fixed threshold defined by a user. However, finding the best threshold is a manual process requiring much effort, and the best one can differ depending on the target data set (or an image). We propose a novel key-point extraction algorithm that automatically adapts to the target image without any manual threshold setting. In our adaptive key-point extraction algorithm, we propose a line-level normalization method to distinguish the lane region from the background clearly. Then, we extract a representative key point for each lane at a line (row of an image) using a kernel density estimation. To check the benefits of our approach, we applied our method to two lane-detection data sets, including TuSimple and CULane. As a result, our method achieved up to 1.80%p and 17.27% better results than using a fixed threshold in the perspectives of accuracy and distance error between the ground truth key-point and the predicted point.

• Ocean Systems Engineering
• /
• v.2 no.2
• /
• pp.137-145
• /
• 2012

Recent developments in underwater image recognition methods have received large attention by the ocean engineering researchers. In this paper, an improved bi-dimensional empirical mode decomposition (BEMD) approach is employed to decompose the given underwater image into intrinsic mode functions (IMFs) and residual. We developed a joint algorithm based on BEMD and Canny operator to extract multi-pixel edge features at multiple scales in IMFs sub-images. So the multiple pixel edge extraction is an advantage of our approach; the other contribution of this method is the realization of the bi-dimensional sifting process, which is realized utilizing regional-based operators to detect local extreme points and constructing radial basis function for curve surface interpolation. The performance of the multi-pixel edge extraction algorithm for processing underwater image is demonstrated in the contrast experiment with both the proposed method and the phase congruency edge detection.

• Journal of Institute of Control, Robotics and Systems
• /
• v.15 no.5
• /
• pp.488-495
• /
• 2009

This paper presents a fast feature extraction method for autonomous mobile robots utilizing parallel processing and based on OpenMP, SSE (Streaming SIMD Extension) and CUDA programming. In the first step on CPU version, the algorithms and codes are optimized and then implemented by parallel processing. The parallel algorithms are debugged to maintain the same level of performance and the process for extracting key points and obtaining dominant orientation with respect to key points is parallelized. After extraction, a parallel descriptor via SSE instructions is constructed. And the GPU version also implemented by parallel processing using CUDA based on the SIFT. The GPU-Parallel descriptor achieves an acceleration up to five times compared with the CPU-Parallel descriptor, but it shows the lower performance than CPU version. CPU version also speed-up the four and half times compared with the original SIFT while maintaining robust performance.

• Journal of Electrical Engineering and Technology
• /
• v.12 no.1
• /
• pp.363-372
• /
• 2017

This paper presents a novel face recognition algorithm based on the deep convolution neural network and key point detection jointed local binary pattern methodology to enhance the accuracy of face recognition. We firstly propose the modified face key feature point location detection method to enhance the traditional localization algorithm to better pre-process the original face images. We put forward the grey information and the color information with combination of a composite model of local information. Then, we optimize the multi-layer network structure deep learning algorithm using the Fisher criterion as reference to adjust the network structure more accurately. Furthermore, we modify the local binary pattern texture description operator and combine it with the neural network to overcome drawbacks that deep neural network could not learn to face image and the local characteristics. Simulation results demonstrate that the proposed algorithm obtains stronger robustness and feasibility compared with the other state-of-the-art algorithms. The proposed algorithm also provides the novel paradigm for the application of deep learning in the field of face recognition which sets the milestone for further research.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.37 no.2
• /
• pp.55-62
• /
• 2019

The stereo geometry establishment based on the precise sensor modeling is prerequisite for accurate stereo data processing. Ground control points are generally required for the accurate sensor modeling though it is not possible over the area where the accessibility is limited or reference data is not available. For the areas, the relative orientation should be carried out to improve the geometric consistency between the stereo data though it does not improve the absolute positional accuracy. The relative orientation requires conjugate points that are well distributed over the entire image region. Therefore the automatic conjugate point extraction is required because the manual operation is labor-intensive. In this study, we applied the method consisting of the key point extraction, the search space minimization based on the epipolar line, and the rigorous outlier detection based on the RPCs (Rational Polynomial Coefficients) bias compensation modeling. We tested different parameters of window sizes for Kompsat-2 across track stereo data and analyzed the RPCs precision after the bias compensation for the cases whether the epipolar line information is used or not. The experimental results showed that matching outliers were inevitable for the different matching parameterization but they were successfully detected and removed with the rigorous method for sub-pixel level of stereo RPCs precision.

• Journal of the Semiconductor & Display Technology
• /
• v.12 no.3
• /
• pp.59-64
• /
• 2013

Accurate and robust image registration is important task in many applications such as image retrieval and computer vision. To perform the image registration, essential required steps are needed in the process: feature detection, extraction, matching, and reconstruction of image. In the process of these function, feature extraction not only plays a key role, but also have a big effect on its performance. There are two representative algorithms for extracting image features, which are scale invariant feature transform (SIFT) and speeded up robust feature (SURF). In this paper, we present and evaluate two methods, focusing on comparative analysis of the performance. Experiments for accurate and robust feature detection are shown on various environments such like scale changes, rotation and affine transformation. Experimental trials revealed that SURF algorithm exhibited a significant result in both extracting feature points and matching time, compared to SIFT method.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.26 no.4
• /
• pp.407-414
• /
• 2008

This study introduces an automatic image matching algorithm that can be applied for the scale different image pairs consisting of the satellite pushbroom images and the aerial frame images. The proposed method is based on several image processing techniques such as pre-processing, filtering, edge thinning, interest point extraction, and key-descriptor matching, in order to enhance the matching accuracy and the processing speed. The proposed method utilizes various characteristics, such as the different geometry of image acquisition and the different radiometric characteristics, of the multi-sensor images. In addition, the suggested method uses the sensor model to minimize search area and eliminate false-matching points automatically.

• Proceedings of the Korean Society of Surveying, Geodesy, Photogrammetry, and Cartography Conference
• /
• 2004.04a
• /
• pp.171-176
• /
• 2004

Image matching is fundamental process in photogrammetry and computer vision to identify and to measure corresponding features on the multiple images. Uniqueness of the matching entities and robustness of the algorithm are the key issues that have influence on quality of the matching result. The optimal solution could be obtained by utilizing appropriate matching entities in the first place. In this study, candidate matching points were extracted by interest operator, and an area-based matching method was applied with characteristics of the gray value distribution as the matching entities. The characteristic information is based on the concept of "intrinsic image" (or parameter image). The information was utilized as additional and/or complementary matching entities. Matching on interest points with the characteristic information resulted in high quality of matching because matching windows were created with surrounding pixels of the interest points that contain distinct and unique features. The experiment shows that matching quality and reliability increase by exploiting interest operator, and the characteristic information has potential to be matching entity.