• The KIPS Transactions:PartD
• /
• v.16D no.3
• /
• pp.451-462
• /
• 2009

Collaborative Filtering (CF), one of the most successful technologies among recommender systems, is a system assisting users in easily finding the useful information. However, despite its success and popularity, CF encounters a serious limitation with quality evaluation, called cold start problems. To alleviate this limitation, in this paper, we propose a unique method of building models derived from explicit ratings and applying the models to CF recommender systems. The proposed method is divided into two phases, an offline phase and an online phase. First, the offline phase is a building pre-computed model phase in which most of tasks can be conducted. Second, the online phase is either a prediction or recommendation phase in which the models are used. In a model building phase, we first determine a priori predicted rating and subsequently identify prediction errors for each user. From this error information, an error-reflected model is constructed. The error-reflected model, which is reflected average prior prediction errors of user neighbors and item neighbors, can make accurate predictions in the situation where users or items have few opinions; this is known as the cold start problems. In addition, in order to reduce the re-building tasks, the error-reflected model is designed such that the model is updated effectively and users'new opinions are reflected incrementally, even when users present a new rating feedback.

• Korean Journal of Remote Sensing
• /
• v.33 no.4
• /
• pp.445-454
• /
• 2017

The performance of spatial downscaling models depends on the quality of input coarse scale products. Thus, the impact of intrinsic errors contained in coarse scale satellite products on predictive performance should be properly assessed in parallel with the development of advanced downscaling models. Such an assessment is the main objective of this paper. Based on a synthetic satellite precipitation product at a coarse scale generated from rain gauge data, two synthetic precipitation products with different amounts of error were generated and used as inputs for spatial downscaling. Geographically weighted regression, which typically has very high explanatory power, was selected as the trend component estimation model, and area-to-point kriging was applied for residual correction in the spatial downscaling experiment. When errors in the coarse scale product were greater, the trend component estimates were much more susceptible to errors. But residual correction could reduce the impact of the erroneous trend component estimates, which improved the predictive performance. However, residual correction could not improve predictive performance significantly when substantial errors were contained in the input coarse scale data. Therefore, the development of advanced spatial downscaling models should be focused on correction of intrinsic errors in the coarse scale satellite product if a priori error information could be available, rather than on the application of advanced regression models with high explanatory power.

• The Transactions of the Korea Information Processing Society
• /
• v.7 no.8
• /
• pp.2555-2564
• /
• 2000

In this paper, a new unsupervised image segmentation algorithm is proposed. To model the contextual information presented in images, the characteristics of the Markov random fields (MRF) are utilized. Textured images are modeled as realizations of the stationary Gaussian MRF on a two-dimensional square lattice using the conditional autoregressive (CAR) equations with a second-order noncausal neighborhood. To detect boundaries, hypothesis tests over two masked areas are performed. Under the hypothesis, masked areas are assumed to belong to the same class of textures and CAR equation parameters are estimated in a minimum-mean-square-error (MMSE) sense. If the hypothesis is rejected, a measure of dissimilarity between two areas is accumulated on the rejected area. This approach produces potential edge maps. Using these maps, boundary detection can be performed, which resulting no micro edges. The performance of the proposed algorithm is evaluated by some experiments using real images as weB as synthetic ones. The experiments demonstrate that the proposed algorithm can produce satisfactorY segmentation without any a priori information.

• Journal of the Korean Institute of Telematics and Electronics S
• /
• v.35S no.12
• /
• pp.55-62
• /
• 1998

Matching two images is an essential step for many computer vision applications. A new approach to the scale and rotation invariant scene matching, using linear features, is presented. Scene or model images are described by a set of linear features approximating edge information, which can be obtained by the conventional edge detection, thinning, and piecewise linear approximation. A set of candidate parameters are hypothesized by mapping the angular difference and a new distance measure to the Hough space and by detecting maximally consistent points. These hypotheses are verified by a fast linear feature matching algorithm composed of a single-step relaxation and a Hough technique. The proposed method is shown to be much faster than the conventional one where the relaxation process is repeated until convergence, while providing matching performance robust to the random alteration of the linear features, without a priori information on the geometrical transformation parameters.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• v.12 no.1
• /
• pp.413-435
• /
• 2018

Although the accuracy of handwritten character recognition based on deep networks has been shown to be superior to that of the traditional method, the use of an overly deep network significantly increases time consumption during parameter training. For this reason, this paper took the training time and recognition accuracy into consideration and proposed a novel handwritten character recognition algorithm with newly designed network structure, which is based on an extended nonlinear kernel residual network. This network is a non-extremely deep network, and its main design is as follows:(1) Design of an unsupervised apriori algorithm for intra-class clustering, making the subsequent network training more pertinent; (2) presentation of an intermediate convolution model with a pre-processed width level of 2;(3) presentation of a composite residual structure that designs a multi-level quick link; and (4) addition of a Dropout layer after the parameter optimization. The algorithm shows superior results on MNIST and SVHN dataset, which are two character benchmark recognition datasets, and achieves better recognition accuracy and higher recognition efficiency than other deep structures with the same number of layers.

• Journal of the Institute of Electronics Engineers of Korea SC
• /
• v.42 no.2 s.302
• /
• pp.1-10
• /
• 2005

This paper presents a technique for localization of a mobile robot using a single ultrasonic sensor. The mobile robot is designed for operating in a well-structured environment that can be represented by planes, edges, corners and cylinders in the view of structural features. In the case of ultrasonic sensors, these features have the range information in the form of the arc of a circle that is generally named as RCD (Region of Constant Depth). Localization is the continual provision of a knowledge of position which is deduced from it's a priori position estimation. The environment of a robot is modeled into a two dimensional grid map. we defines a physically-based sonar sensor model and employs an extended Kalman filter to estimate position of the robot. The performance and simplicity of the approach is demonstrated with the results produced by sets of experiments using a mobile robot.

• Investigative Magnetic Resonance Imaging
• /
• v.2 no.1
• /
• pp.58-66
• /
• 1998

Purpose : This paper introduces a new three dimensional magnetic Resonance Image classification which is based on Mar kov Random Field-Gibbs Random Field with a line model. Material and Methods : The performance of the Gibbs Classifier over a statistically heterogeneous image can be improved if the local stationary regions in the image are disassociated from each other through the mechanism of the interaction parameters defined at the local neighborhood level. This usually involves the construction of a line model for the image. In this paper we construct a line model for multisignature images based on the differential of the image which can provide an a priori estimate of the unobservable line field, which may lie in regions with significantly different statistics. the line model estimated from the original image data can in turn be used to alter the values of the interaction parameters of the Gibbs Classifier. Results : MRF-Gibbs classifier for volumetric MR images is developed under the condition that the domain of the image classification is $E^{3}$ space rather thatn the conventional $E^{2}$ space. Compared to context free classification, MRF-Gibbs classifier performed better in homogeneous and along boundaries since contextual information is used during the classification. Conclusion : We construct a line model for multisignature, multidimensional image and derive the interaction parameter for determining the energy function of MRF-Gibbs classifier.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.33B no.12
• /
• pp.71-83
• /
• 1996

This paper describes facial component detection and skew correction algorithm for face recognition. We use a priori knowledge and models about isolated regions to detect eye location from the face image captured in natural office environments. The relations between human face components are represented by several rules. We adopt an edge detection algorithm using sobel mask and 8-connected labelling algorith using array pointers. A labeled image has many isolated components. initially, the eye size rules are used. Eye size rules are not affected much by irregular input image conditions. Eye size rules size, and limited in the ratio between gorizontal and vertical sizes. By the eye size rule, 2 ~ 16 candidate eye components can be detected. Next, candidate eye parirs are verified by the information of location and shape, and one eye pair location is decided using face models about eye and eyebrow. Once we extract eye regions, we connect the center points of the two eyes and calculate the angle between them. Then we rotate the face to compensate for the angle so that the two eyes on a horizontal line. We tested 120 input images form 40 people, and achieved 91.7% success rate using eye size rules and face model. The main reasons of the 8.3% failure are due to components adjacent to eyes such as eyebrows. To detect facial components from the failed images, we are developing a mouth region processing module.

• The Journal of Korean Institute of Information Technology
• /
• v.16 no.11
• /
• pp.97-103
• /
• 2018

In this paper, the estimation method for the power signal harmonics is proposed by using the time-varying optimal finite impulse response (FIR) filter. To estimate the magnitude and phase-angle of the harmonic components, the time-varying optimal FIR filter is designed for the state space representation of the noisy power signal which the magnitude and phase is considered as a stochastic process. Since the time-varying optimal FIR filter used in the proposed method does not use any priori information of the initial condition and has FIR structure, the proposed method could overcome the demerits of Kalman filter based method such as poor estimation and divergence problem. Due to the FIR structure, the proposed method is more robust against to the model uncertainty than the Kalman filter. Moreover, the proposed method gives more general solution than the time-invariant optimal FIR filter based harmonic estimation method. To verify the performance and robustness of the proposed method, the proposed method is compared with time-varying Kalman filter based method through simulation.

• Journal of the Institute of Electronics Engineers of Korea SP
• /
• v.39 no.2
• /
• pp.116-127
• /
• 2002

A new approach to improve the speaker adaptation algorithm by means of the variable number of observation density functions for CDHMM speech recognizer has been proposed. The proposed method uses the observation density function with more than one mixture in each state to represent speech characteristics in detail. The number of mixtures in each state is determined by the number of frames and the determinant of the variance, respectively. The each MAP Parameter is extracted in every mixture determined by these two methods. In addition, the state segmentation method requiring speaker adaptation can segment the adapting speech more Precisely by using speaker-independent model trained from sufficient database as a priori knowledge. And the state duration distribution is used lot adapting the speech duration information owing to speaker's utterance habit and speed. The recognition rate of the proposed methods are significantly higher than that of the conventional method using one mixture in each state.