• Journal of the Semiconductor & Display Technology
• /
• v.12 no.1
• /
• pp.1-7
• /
• 2013

Grey-level statistical models have been widely used in many applications for object location and identification. However, conventional models yield some problems in model refinement when training images are not properly aligned, and have difficulties for real-time recognition of arbitrarily rotated models. This paper presents improved grey-level statistical models that align training images using image or feature matching to overcome problems in model refinement of conventional models, and that enable real-time recognition of arbitrarily rotated objects using efficient hierarchical search methods. Edges or features extracted from a mean training image are used for accurate alignment of models in the search image. On the aligned position and orientation, fitness measure based on grey-level statistical models is computed for object recognition. It is demonstrated in various experiments in PCB inspection that proposed methods are superior to conventional methods in recognition accuracy and speed.

• Journal of Electrical Engineering and Technology
• /
• v.7 no.6
• /
• pp.1001-1008
• /
• 2012

Non-destructive testing (NDT) is a technique used in science and industry to evaluate the properties of a material without causing damage. In this paper we propose a method for segmenting radiographic images of welding in order to extract the welding defects which may occur during the welding process. We study different methods of level set and choose the model adapted to our application. The methods presented here take the property of local segmentation geodesic active contours and have the ability to change the topology automatically. The computation time is considerably reduced after taking into account a new level set function which eliminates the re-initialization procedure. Satisfactory results are obtained after applying this algorithm both on synthetic and real images.

• 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.

• Journal of Korea Multimedia Society
• /
• v.21 no.8
• /
• pp.858-863
• /
• 2018

In this paper, we propose and implement a multi-level convolutional neural network (CNN) algorithm to identify the sexually explicit and lewdness of various image files, and verify its effectiveness by using unauthorized image files generated in the actual military. The proposed algorithm increases the accuracy by applying the convolutional artificial neural network step by step to minimize classification error between similar categories. Experimental data have categorized 20,005 images in the real field into 6 authorization categories and 11 non-authorization categories. Experimental results show that the overall detection rate is 99.51% for the image files. In particular, the excellence of the proposed algorithm is verified through reducing the identification error rate between similar categories by 64.87% compared with the general CNN algorithm.

• Journal of information and communication convergence engineering
• /
• v.10 no.4
• /
• pp.411-419
• /
• 2012

This paper proposes a semi-automatic image-based approach for 3-dimensional (3D) modeling of buildings along streets. Image-based urban 3D modeling techniques are typically based on the use of aerial and ground-level images. The aerial image of the relevant area is extracted from publically available sources in Google Maps by stitching together different patches of the map. Panoramic images are common for ground-level recording because they have advantages for 3D modeling. A panoramic video recorder is used in the proposed approach for recording sequences of ground-level spherical panoramic images. The proposed approach has two advantages. First, detected camera trajectories are more accurate and stable (compared to methods using multi-view planar images only) due to the use of spherical panoramic images. Second, we extract the texture of a facade of a building from a single panoramic image. Thus, there is no need to deal with color blending problems that typically occur when using overlapping textures.

• 한국전산유체공학회:학술대회논문집
• /
• 2011.05a
• /
• pp.259-260
• /
• 2011

Using modern techniques from scientific computing and numerical analysis, natural phenomena or scientific experiment can be simulated effectively with a computer and used for computer graphics, for example as special effects for the film industry, manufacturing the thin film, multi-phase simulation and image processing. The Level Set method can make those things happen without a lot of difficulties. This method was devised by Osher and Sethian(1988) to represent dynamically moving interfaces as the zero level set of a scalar function that evolves in time. Since then, many researchers have worked on many applications using a Level Set Method. I will give a talk about the applications of the Level Set Method.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.28B no.5
• /
• pp.318-326
• /
• 1991

A Quantizer reconstruction level control method for block artifact reduction in DCT image coding is described. In our scheme, quantizer reconstruction level control is obtained by adding quantization level step size to the optimum quantization level in the direction of reducing the block artifact by minimizing the mean square error(MSE) and error difference(EDF) distribution in boundary without the other additional bits. In simulation results, although the performance in terms of signal to noise ratio is degraded by a little amount, mean square of error difference at block boundary and mean square error having relation block artifact is greatly reduced. Subjective image qualities are improved compared with other block artifact reduction method such as postprocessing by filtering and trasform coding by block overlapping. But the addition calculations of 1-dimensional DCT become to be more necessary to coding process for determining the reconstruction level.

• Journal of the Korean Society of Surveying, Geodesy, Photogrammetry and Cartography
• /
• v.24 no.3
• /
• pp.289-297
• /
• 2006

The Land Cover/Land Use Map have been constructed from 1998, which has hierarchical structure according to land cover/land use system. Level 1 classification Map have done using Landsat satellite image over whole Korean peninsula. Level II classification Map have been digitized using IRS-1C, 1D, KOMPSAT and SPOT5 satellite images resolution-merged with low resolution color images. Level II Land Cover/Land Use Map construction by digitizing method, however, is consuming enormous expense for satellite image acquisition, image process and Land Cover/Land Use Map construction. In this paper, the possibility of constructing Level II Land Cover/Land Use Map using hyperspectral satellite image of EO-1 Hyperion, which is studied a lot recently, is studied. The comparison of classifications using Hyperion satellite image offering more spectral information and Landsat-7 ETM+ image is performed to evaluate the availability of Hyperion satellite image. Also, the algorithm of the optimal band selection is presented for effective application of hyperspectral satellite image.

• The KIPS Transactions:PartB
• /
• v.13B no.2 s.105
• /
• pp.121-126
• /
• 2006

In this paper, we propose an approach to increase the power of clustering Web images by using high-level semantic features from text information relevant to Web images as well as low-level visual features of image itself. These high-level text features can be obtained from image URLs and file names, page titles, hyperlinks, and surrounding text. As a clustering engine, self-organizing map (SOM) proposed by Kohonen is used. In the SOM-based clustering using high-level text features and low-level visual features, the 200 images from 10 categories are divided in some suitable clusters effectively. For the evaluation of clustering powers, we propose simple but novel measures indicating the degrees of scattering images from the same category, and degrees of accumulation of the same category images. From the experiment results, we find that the high-level text features are more useful in SOM-based Web image clustering.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.8 no.8
• /
• pp.2895-2912
• /
• 2014

Content-based image retrieval has been the most important technique for managing huge amount of images. The fundamental yet highly challenging problem in this field is how to measure the content-level similarity based on the low-level image features. The primary difficulties lie in the great variance within images, e.g. background, illumination, viewpoint and pose. Intuitively, an ideal similarity measure should be able to adapt the data distribution, discover and highlight the content-level information, and be robust to those variances. Motivated by these observations, we in this paper propose a probabilistic similarity learning approach. We first model the distribution of low-level image features and derive the free energy kernel (FEK), i.e., similarity measure, based on the distribution. Then, we propose a learning approach for the derived kernel, under the criterion that the kernel outputs high similarity for those images sharing the same class labels and output low similarity for those without the same label. The advantages of the proposed approach, in comparison with previous approaches, are threefold. (1) With the ability inherited from probabilistic models, the similarity measure can well adapt to data distribution. (2) Benefitting from the content-level hidden variables within the probabilistic models, the similarity measure is able to capture content-level cues. (3) It fully exploits class label in the supervised learning procedure. The proposed approach is extensively evaluated on two well-known databases. It achieves highly competitive performance on most experiments, which validates its advantages.