• Journal of the Institute of Electronics and Information Engineers
• /
• v.50 no.8
• /
• pp.187-195
• /
• 2013

Most of former studies for car license plate detection restrict the image acquisition environment. The aim of this research is to diminish the restrictions by proposing a new method of using SIFT and neural network. SIFT can be used in diverse situations with less restriction because it provides size- and rotation-invariance and large discriminating power. SIFT extracted from the license plate image is divided into the internal(inside class) and the external(outside class) ones and the classifier is trained using them. In the proposed method, by just putting the various types of license plates, the trained neural network classifier can process all of the types. Although the classification performance is not high, the inside class appears densely over the plate region and sparsely over the non-plate regions. These characteristics create a local feature map, from which we can identify the location with the global maximum value as a candidate of license plate region. We collected image database with much less restriction than the conventional researches. The experiment and evaluation were done using this database. In terms of classification accuracy of SIFT keypoints, the correct recognition rate was 97.1%. The precision rate was 62.0% and recall rate was 50.2%. In terms of license plate detection rate, the correct recognition rate was 98.6%.

• Journal of KIISE:Computer Systems and Theory
• /
• v.36 no.5
• /
• pp.404-411
• /
• 2009

We introduce in this paper a new method for smooth foldover-free warping of images, based on the vector field deformation technique proposed by Von Funck et al. It allows users to specify the constraints in two different ways: positional constraints to constrain the position of a point in the image and gradient constraints to constrain the orientation and scaling of some parts of the image. From the user-specified constraints, it computes in the image domain a C1-continuous velocity vector field, along which each pixel progressively moves from its original position to the target. The target positions of the pixels are obtained by solving a set of partial derivative equations with the 4th order Runge-Kutta method. We show how our method can be useful for texture mapping with hard constraints. We start with an unconstrained planar embedding of a target mesh using a previously known method (Least Squares Conformal Map). Then, in order to obtain a texture map that satisfies the given constraints, we use the proposed warping method to align the features of the texture image with those on the unconstrained embedding. Compared to previous work, our method generates a smoother texture mapping, offers higher level of control for defining the constraints, and is simpler to implement.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.42 no.4 s.304
• /
• pp.95-102
• /
• 2005

In this paper, we propose a user friendly object-based image retrieval system using the interaction between cortex and hippocampus. Most existing ways of queries in content-based image retrieval rely on query by example or query by sketch. But these methods of queries are not adequate to needs of people's various queries because they are not easy for people to use and restrict. We propose a method of automatic color object extraction using CSB tree map(Color and Spatial based Binary をn map). Extracted objects were transformed to bit stream representing information such as color, size and location by region labelling algorithm and they are learned by the hippocampal neural network using the interplay between cortex and hippocampus. The cells of exciting at peculiar features in brain generate the special sign when people recognize some patterns. The existing neural networks treat each attribute of features evenly. Proposed hippocampal neural network makes an adaptive fast content-based image retrieval system using excitatory learning method that forwards important features to long-term memories and inhibitory teaming method that forwards unimportant features to short-term memories controlled by impression.

• The Journal of Korea Institute of Information, Electronics, and Communication Technology
• /
• v.12 no.5
• /
• pp.512-520
• /
• 2019

It is quite useful to represent speeches visually for learners who study foreign languages as well as the hearing impaired who cannot directly hear speeches, and a number of researches have been presented in the literature. They remain, however, at the level of representing the characteristics of speeches using colors or showing the changing shape of lips and mouth using the animation-based representation. As a result of such approaches, those methods cannot tell the users how far their pronunciations are away from the standard ones, and moreover they make it technically difficult to develop such a system in which users can correct their pronunciation in an interactive manner. In order to address these kind of drawbacks, this paper proposes a speech visualization model based on the relative distance between the user's speech and the standard one, furthermore suggests actual implementation directions by applying the proposed model to the visualization of Korean vowels. The method extract three formants F1, F2, and F3 from speech signals and feed them into the Kohonen's SOM to map the results into 2-D screen and represent each speech as a pint on the screen. We have presented a real system implemented using the open source formant analysis software on the speech of a Korean instructor and several foreign students studying Korean language, in which the user interface was built using the Javascript for the screen display.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.2
• /
• pp.33-42
• /
• 2022

In this paper, we proposed a deep learning model based on multi-scale dilated convolution feature fusion for the segmentation of black ice region in road image to send black ice warning to drivers in real time. In the proposed multi-scale dilated convolution feature fusion network, different dilated ratio convolutions are connected in parallel in the encoder blocks, and different dilated ratios are used in different resolution feature maps, and multi-layer feature information are fused together. The multi-scale dilated convolution feature fusion improves the performance by diversifying and expending the receptive field of the network and by preserving detailed space information and enhancing the effectiveness of diated convolutions. The performance of the proposed network model was gradually improved with the increase of the number of dilated convolution branch. The mIoU value of the proposed method is 96.46%, which was higher than the existing networks such as U-Net, FCN, PSPNet, ENet, LinkNet. The parameter was 1,858K, which was 6 times smaller than the existing LinkNet model. From the experimental results of Jetson Nano, the FPS of the proposed method was 3.63, which can realize segmentation of black ice field in real time.

• Journal of Digital Convergence
• /
• v.13 no.11
• /
• pp.331-336
• /
• 2015

The texture mapping used for the low-polygon models is one of the important workflows in the graphical representation of the 3D game. Only one hand painted texture is mapped on the surface of the 3D model and represents the color of the material and visual sense of touching by itself in that process. In the 3D game graphics, it is very important to visualize the textile sensation such as protruding and denting. It can be interpreted by the Gestalt Law to recognize a plane as a 3D sense of volume. Moreover, the concept of Affordance is necessary to recognize and perceive the textile sensation. It means visual recognizing of that relationship in the learning process. In this paper, The questionnaire survey targeting 3D game graphic designers is carried out. By analyzing the survey results, we suggest the important characteristic in the process of making hand painted texture.

• Journal of Digital Contents Society
• /
• v.18 no.8
• /
• pp.1635-1640
• /
• 2017

In this paper, a hierarchical image tree model for weather classification is defined in a single outdoor image, and a weather classification algorithm using image intensity and k-mean segmentation image is proposed. In the first level of the hierarchical image tree model, the indoor and outdoor images are distinguished. Whether the outdoor image is daytime, night, or sunrise/sunset image is judged using the intensity and the k-means segmentation image at the second level. In the last level, if it is classified as daytime image at the second level, it is finally estimated whether it is sunny or foggy image based on edge map and fog rate. Some experiments are conducted so as to verify the weather classification, and as a result, the proposed method shows that weather features are effectively detected in a given image.

• Korean Journal of Remote Sensing
• /
• v.37 no.5_1
• /
• pp.1061-1070
• /
• 2021

In this manuscript, we tried to improve the performance of the FC-DenseNet by applying an attention gate for the classification of cropping areas. The attention gate module could facilitate the learning of a deep learning model and improve the performance of the model by injecting of spatial/spectral weights to each feature map. Crop classification was performed in the onion and garlic regions using a proposed deep learning model in which an attention gate was added to the skip connection part of FC-DenseNet. Training data was produced using various PlanetScope satellite imagery, and preprocessing was applied to minimize the problem of imbalanced training dataset. As a result of the crop classification, it was verified that the proposed deep learning model can more effectively classify the onion and garlic regions than existing FC-DenseNet algorithm.

• Journal of Broadcast Engineering
• /
• v.26 no.5
• /
• pp.608-620
• /
• 2021

Recent deep learning-based face super-resolution (FSR) works showed significant performances by utilizing facial prior knowledge such as facial landmark and dictionary that reflects structural or semantic characteristics of the human face. However, most of these methods require additional processing time and memory. To solve this issue, this paper propose an efficient FSR models using knowledge distillation techniques. The intermediate features of teacher network which contains dictionary information based on major face regions are transferred to the student through adversarial multi-scale features distillation. Experimental results show that the proposed model is superior to other SR methods, and its effectiveness compare to teacher model.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.24 no.12
• /
• pp.1604-1611
• /
• 2020

Light field image captured by a microlens array-based camera has many limitations in practical use due to its low spatial resolution and angular resolution. High spatial resolution images can be easily acquired with a single image super-resolution technique that has been studied a lot recently. But there is a problem in that high angular resolution images are distorted in the process of using disparity information inherent among images, and thus it is difficult to obtain a high-quality angular resolution image. In this paper, we propose light field angular super-resolution that extracts an initial feature map using an dilated convolutional neural network in order to effectively extract the view difference information inherent among images and generates target image using a residual neural network. The proposed network showed superior performance in PSNR and subjective image quality compared to existing angular super-resolution networks.