• Journal of Digital Contents Society
• /
• v.18 no.1
• /
• pp.183-190
• /
• 2017

This paper proposes an audio event detection method using Deep Neural Networks (DNN). The proposed method applies Feed Forward Neural Network (FFNN) to generate output probabilities of twenty audio events for each frame. Mel scale filter bank (FBANK) features are extracted from each frame, and its five consecutive frames are combined as one vector which is the input feature of the FFNN. The output layer of FFNN produces audio event probabilities for each input feature vector. More than five consecutive frames of which event probability exceeds threshold are detected as an audio event. An audio event continues until the event is detected within one second. The proposed method achieves as 71.8% accuracy for 20 classes of the UrbanSound8K and the BBC Sound FX dataset.

• Journal of the Korea Society of Computer and Information
• /
• v.11 no.6 s.44
• /
• pp.269-277
• /
• 2006

We propose a new expert system for recovering the broken fragments of relics into an original form using computer graphics and image processing. This paper presents a system with an application to tombstones objects of flat plane with letters carved in for assembling the fragments by placing their respective fragments in the right position. The matching process contains three sub-processes: aligning the front and letters of an object, identifying the matching directions, and determining the detailed matching positions. We apply least squares fitting, vector inner product, and geometric and RGB errors to the matching process. It turned out that 2-D translations via fragments-alignment enable us to save the computational load significantly. Based on experimental results from the damaged cultural fragments, the performance of the proposed method is illustrated.

• The KIPS Transactions:PartB
• /
• v.8B no.5
• /
• pp.515-522
• /
• 2001

Feature-based similarity retrieval become an important research issue in image database systems. The features of image data are useful to discrimination of images. In this paper, we propose the highspeed k-Nearest Neighbor search algorithm based on Self-Organizing Maps. Self-Organizing Map(SOM) provides a mapping from high dimensional feature vectors onto a two-dimensional space. A topological feature map preserves the mutual relations (similarity) in feature spaces of input data, and clusters mutually similar feature vectors in a neighboring nodes. Each node of the topological feature map holds a node vector and similar images that is closest to each node vector. We implemented about k-NN search for similar image classification as to (1) access to topological feature map, and (2) apply to pruning strategy of high speed search. We experiment on the performance of our algorithm using color feature vectors extracted from images. Promising results have been obtained in experiments.

• Aerospace Engineering and Technology
• /
• v.3 no.2
• /
• pp.56-64
• /
• 2004

In this research, design results of computer hardware which control solid motor movable nozzle thrust vector control(TVC) actuator for Korea Space Launch Vehicle I(KSLV-I) are described. TVC computer hardware is the equipment which has jobs for receiving control commands from Navigation Guidance Unit(NGU) and then actuating TVC actuator. Also, it has ability to communicate with other on board or ground equipments. Computer hardware has a digital signal processor as the main processor which is capable of high speed calculating ability of control algorithm, so it can have more stability, reliability and flexibility than the previous analog controller of KSR-III. Target board was designed for on board program development and then first prototype hardware was developed. Top level system design criteria, hardware configurations and ground support equipment of TVC computer system are described.

• Journal of Institute of Control, Robotics and Systems
• /
• v.21 no.10
• /
• pp.965-971
• /
• 2015

This paper suggests a feature point-based Iterative Closest Point (ICP) algorithm to compensate for the disparity error in building a two-dimensional map. The ICP algorithm is a typical algorithm for matching a common object in two different images. In the process of building a two-dimensional map using the laser scanner data, warping and distortions exist in the map because of the disparity between the two sensor values. The ICP algorithm has been utilized to reduce the disparity error in matching the scanned line data. For this matching process in the conventional ICP algorithm, pre-known reference data are required. Since the proposed algorithm extracts characteristic points from laser-scanned data, reference data are not required for the matching. The laser scanner starts from the right side of the mobile robot and ends at the left side, which causes disparity in the scanned line data. By finding the matching points between two consecutive frame images, the motion vector of the mobile robot can be obtained. Therefore, the disparity error can be minimized by compensating for the motion vector caused by the mobile robot motion. The validity of the proposed algorithm has been verified by comparing the proposed algorithm in terms of map-building accuracy to conventional ICP algorithm real experiments.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.36S no.8
• /
• pp.101-112
• /
• 1999

In this paper, the codebook design method of VQ (vector quantization) is proposed an method to extract feature data of image for multimedia information searching. Conventional VQ codebook design methods are unsuitable to extract the feature data of images because they have too much computation time, memory for vector decoding and blocking effects like DCT (discrete cosine transform). The proposed design method is consists of the feature extraction by WT (wavelet transform) and the data group divide method by PCA (principal component analysis). WT is introduced to remove the blocking effect of an image with high compressing ratio. Computer simulations show that the proposed method has the better performance in processing speed than the VQ design method using SOM (self-organizing map).

• Proceedings of the Korean Institute of Information and Commucation Sciences Conference
• /
• 2016.05a
• /
• pp.101-103
• /
• 2016

Recently, the target object can be represented as sparse coefficient vector in visual tracking. Due to this reason, exploitation of the compressibility in the transform domain by using L1 minimization is needed. Further, L1 minimization is proposed to handle the occlusion problem in visual tracking, since tracking failures mostly are caused by occlusion. Furthermore, there is a weighted parameter in L1 minimization that influences the result of this minimization. In this paper, this parameter is analyzed for occlusion problem in visual tracking. Several coefficients that derived from median value of the target object, mean value of the arget object, the standard deviation of the target object are, 0, 0.1, and 0.01 are used as weighted parameter of L1 minimization. Based on the experimental results, the value which is equal to 0.1 is suggested as weighted parameter of L1 minimization, due to achieved the best result of success rate and precision performance parameter. Both of these performance parameters are based on one pass evaluation (OPE).

• Journal of the Korea Society of Computer and Information
• /
• v.17 no.12
• /
• pp.101-109
• /
• 2012

In many diagnostic and treatment procedures for prostate disease accurate detection of prostate boundaries in transrectal ultrasound(TRUS) images is required. This is a challenging and difficult task due to weak prostate boundaries, speckle noise and the short range of gray levels. In this paper a method for automatic prostate segmentation inTRUS images using support vectors and snake-like contour is presented. This method involves preprocessing, extracting Gabor feature, training, and prostate segmentation. Gabor filter bank for extracting the texture features has been implemented. A support vector machine(SVM) for training step has been used to get each feature of prostate and nonprostate. The boundary of prostate is extracted by the snake-like contour algorithm. The results showed that this new algorithm extracted the prostate boundary with less than 9.3% relative to boundary provided manually by experts.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.15 no.10
• /
• pp.3708-3728
• /
• 2021

Indian herbal plants are used in agriculture and in the food, cosmetics, and pharmaceutical industries. Laboratory-based tests are routinely used to identify and classify similar herb species by analyzing their internal cell structures. In this paper, we have applied computer vision techniques to do the same. The original leaf image was preprocessed using the Chan-Vese active contour segmentation algorithm to efface the background from the image by setting the contraction bias as (v) -1 and smoothing factor (µ) as 0.5, and bringing the initial contour close to the image boundary. Thereafter the segmented grayscale image was fed to a leaky capacitance fired neuron model (LCFN), which differentiates between similar herbs by combining different groups of pixels in the leaf image. The LFCN's decay constant (f), decay constant (g) and threshold (h) parameters were empirically assigned as 0.7, 0.6 and h=18 to generate the 1D feature vector. The LCFN time sequence identified the internal leaf structure at different iterations. Our proposed framework was tested against newly collected herbal species of natural images, geometrically variant images in terms of size, orientation and position. The 1D sequence and shape features of aloe, betel, Indian borage, bittergourd, grape, insulin herb, guava, mango, nilavembu, nithiyakalyani, sweet basil and pomegranate were fed into the 5-fold Bayesian regularization neural network (BRNN), K-nearest neighbors (KNN), support vector machine (SVM), and ensemble classifier to obtain the highest classification accuracy of 91.19%.

• Journal of the Korea Society of Computer and Information
• /
• v.27 no.5
• /
• pp.21-28
• /
• 2022

All health problems that occur in the circulatory system are refer to cardiovascular illness, such as heart and vascular diseases. Deaths from cardiovascular disorders are recorded one third of in total deaths in 2019 worldwide, and the number of deaths continues to rise. Therefore, if it is possible to predict diseases that has high mortality rate with patient's data and AI system, they would enable them to be detected and be treated in advance. In this study, models are produced to predict heart disease, which is one of the cardiovascular diseases, and compare the performance of models with Accuracy, Precision, and Recall, with description of the way of improving the performance of the Decision Tree(Decision Tree, KNN (K-Nearest Neighbor), SVM (Support Vector Machine), and DNN (Deep Neural Network) are used in this study.). Experiments were conducted using scikit-learn, Keras, and TensorFlow libraries using Python as Jupyter Notebook in macOS Big Sur. As a result of comparing the performance of the models, the Decision Tree demonstrates the highest performance, thus, it is recommended to use the Decision Tree in this study.