• The KIPS Transactions:PartB
• /
• v.10B no.6
• /
• pp.657-664
• /
• 2003

Reliable tracking of moving humans is essential to motion estimation, video surveillance and human-computer interface. This paper presents a new approach to human motion tracking that combines appearance-based and model-based techniques. Monocular color video is processed at both pixel level and object level. At the pixel level, a Gaussian mixture model is used to train and classily individual pixel colors. At the object level, a 3D human body model projected on a 2D image plane is used to fit the image data. Our method does not use inverse kinematics due to the singularity problem. While many others use stochastic sampling for model-based motion tracking, our method is purely dependent on nonlinear programming. We convert the human motion tracking problem into a nonlinear programming problem. A cost function for parameter optimization is used to estimate the degree of the overlapping between the foreground input image silhouette and a projected 3D model body silhouette. The overlapping is computed using computational geometry by converting a set of pixels from the image domain to a polygon in the real projection plane domain. Our method is used to recognize various human motions. Motion tracking results from video sequences are very encouraging.

• Smart Media Journal
• /
• v.12 no.9
• /
• pp.45-59
• /
• 2023

Weakly supervised object localization (WSOL) is a task of localizing an object in an image using only image-level labels. Previous studies have followed the conventional class activation mapping (CAM) pipeline. However, we reveal the current CAM approach suffers from problems which cause original CAM could not capture the complete defects features. This work utilizes a convolutional neural network (CNN) pretrained on image-level labels to generate class activation maps in a multi-scale manner to highlight discriminative regions. Additionally, a vision transformer (ViT) pretrained was treated to produce multi-head attention maps as an auxiliary detector. By integrating the CNN-based CAMs and attention maps, our approach localizes defective regions without requiring bounding box or pixel-level supervision during training. We evaluate our approach on a dataset of apple images with only image-level labels of defect categories. Experiments demonstrate our proposed method aligns with several Object Detection models performance, hold a promise for improving localization.

• The KIPS Transactions:PartB
• /
• v.18B no.3
• /
• pp.131-138
• /
• 2011

In this paper, we compare our proposed method with previous methods for the volumetric image segmentation using level set. In order to obtain an exact segmentation, the region and boundary information of image object are used in our proposed speed function. The boundary information is defined by the gradient vector flow obtained from the gradient images and the region information is defined by Gaussian distribution information of pixel intensity in a region-of-interest for image segmentation. Also the regular term is used to remove the noise around surface. We show various experimental results of real medical volume images to verify the superiority of proposed method.

• Korean Journal of Remote Sensing
• /
• v.19 no.4
• /
• pp.329-339
• /
• 2003

In this study, an approach of image fusion in decision level has been proposed for unsupervised image classification using the images acquired from multiple sensors with different characteristics. The proposed method applies separately for each sensor the unsupervised image classification scheme based on spatial region growing segmentation, which makes use of hierarchical clustering, and computes iteratively the maximum likelihood estimates of fuzzy class vectors for the segmented regions by EM(expected maximization) algorithm. The fuzzy class vector is considered as an indicator vector whose elements represent the probabilities that the region belongs to the classes existed. Then, it combines the classification results of each sensor using the fuzzy class vectors. This approach does not require such a high precision in spatial coregistration between the images of different sensors as the image fusion scheme of pixel level does. In this study, the proposed method has been applied to multispectral SPOT and AIRSAR data observed over north-eastern area of Jeollabuk-do, and the experimental results show that it provides more correct information for the classification than the scheme using an augmented vector technique, which is the most conventional approach of image fusion in pixel level.

• The Journal of Korean Institute of Communications and Information Sciences
• /
• v.28 no.9C
• /
• pp.868-875
• /
• 2003

In this paper, enhanced multiresolution motion estimation(MRME) using reduction of one-pixel shift in wavelet domain is proposed. Conventional multiresolution motion estimation using hierarchical relationship of wavelet coefficient has difficulty for accurate motion estimation due to shift-variant property by decimation process of the wavelet transform. Therefore, to overcome shift-variant property of wavelet coefficient, two level wavelet transform is performed. In order too reduce one-pixel shift on low band signal, S$_4$ band is interpolated by inserting average value. Secondly, one level wavelet transform is applied to the interpolated S$_4$ band. To estimate initial motion vector, block matching algorithm is applied to low band signal S$_{8}$. Multiresolution motion estimation is performed at the rest subbands in low level. According to the experimental results, proposed method showed 1-2dB improvement of PSNR performance at the same bit rate as well as subjective quality compared with the conventional multiresolution motion estimation(MRME) methods and full-search block matching in wavelet domain.

• Tunnel and Underground Space
• /
• v.29 no.3
• /
• pp.184-196
• /
• 2019

This study aims to extract a 3D image of micro-cracks generated by hydraulic fracturing tests, using the deep learning method and X-ray computed tomography images. The pixel-level cracks are difficult to be detected via conventional image processing methods, such as global thresholding, canny edge detection, and the region growing method. Thus, the convolutional neural network-based encoder-decoder network is adapted to extract and analyze the micro-crack quantitatively. The number of training data can be acquired by dividing, rotating, and flipping images and the optimum combination for the image augmentation method is verified. Application of the optimal image augmentation method shows enhanced performance for not only the validation dataset but also the test dataset. In addition, the influence of the original number of training data to the performance of the deep learning-based neural network is confirmed, and it leads to succeed the pixel-level crack detection.

• Journal of Institute of Control, Robotics and Systems
• /
• v.17 no.1
• /
• pp.17-21
• /
• 2011

This paper presents a real- time detection of on-road succeeding vehicles based on low level edge features and a boosted cascade of Haar-like features. At first, the candidate vehicle location in an image is found by low level horizontal edge and symmetry characteristic of vehicle. Then a boosted cascade of the Haar-like features is applied to the initial hypothesized vehicle location to extract the refined vehicle location. The initial hypothesis generation using simple edge features speeds up the whole detection process and the application of a trained cascade on the hypothesized location increases the accuracy of the detection process. Experimental results on real world road scenario with processing speed of up to 27 frames per second for $720{\times}480$ pixel images are presented.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.53 no.10
• /
• pp.10-14
• /
• 2016

Holographic data storage (HDS) features short access times, high storage capacities, and fast transfer rates since the data is recorded and read not by lines but by pages within a volume of holographic material. Furthermore, a single pixel can store more than 1 bit if it is multi-level. However, there is a problem of inter-symbol interference (ISI) between the adjacent symbols if the level difference between neighboring symbols is large. Hence, one should avoid side by side placement of the smallest level symbol and the largest level symbol in any direction. This paper proposes a 12/16 modulation code for 4-level holographic data storage, so that the largest symbol is 3 and the smallest symbol is 0, in order to reduce the ISI.

• Journal of the Korean Association of Geographic Information Studies
• /
• v.8 no.4
• /
• pp.14-23
• /
• 2005

This paper proposes a method for matching two different images, especially satellite images. In the general image matching fields, when an image is compared to other image, they may have different properties on the size, contents, brightness, etc. If there is no noise in each image, in other words, they have identical pixel level and unchanged edges, the image matching method will be simple comparison between two images with pixel by pixel. However, in many applications, most of images to be matched should have much different properties. This paper proposes an efficient method for matching satellite images. This method is to match a raw satellite image with GCP chips. From this we can make a geometrically corrected image. The proposed method is based on wavelet transformation, not required any pre-processing such as histogram equalization, analysis of raw image like the traditional methods.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.10
• /
• pp.3475-3489
• /
• 2014

Here we present a simple flame detection method for an infrared (IR) thermal camera based real-time fire surveillance digital signal processor (DSP) system. Infrared thermal cameras are especially advantageous for unattended fire surveillance. All-weather monitoring is possible, regardless of illumination and climate conditions, and the data quantity to be processed is one-third that of color videos. Conventional IR camera-based fire detection methods used mainly pixel-based temporal correlation functions. In the temporal correlation function-based methods, temporal changes in pixel intensity generated by the irregular motion and spreading of the flame pixels are measured using correlation functions. The correlation values of non-flame regions are uniform, but the flame regions have irregular temporal correlation values. To satisfy the requirement of early detection, all fire detection techniques should be practically applied within a very short period of time. The conventional pixel-based correlation function is computationally intensive. In this paper, we propose an IR camera-based simple flame detection algorithm optimized with a compact embedded DSP system to achieve early detection. To reduce the computational load, block-based calculations are used to select the candidate flame region and measure the temporal motion of flames. These functions are used together to obtain the early flame detection algorithm. The proposed simple algorithm was tested to verify the required function and performance in real-time using IR test videos and a real-time DSP system. The findings indicated that the system detected the flames within 5 to 20 seconds, and had a correct flame detection ratio of 100% with an acceptable false detection ratio in video sequence level.