• Proceedings of the Korea Information Processing Society Conference
• /
• 2023.05a
• /
• pp.687-688
• /
• 2023

Bone age assessment is a crucial task in pediatric radiology for assessing growth and development in children. In this paper, we explore the potential of Vision Transformer, a state-of-the-art deep learning model, for bone age assessment using X-ray images. We generate heatmap outputs using a pre-trained Vision Transformer model on a publicly available dataset of hand X-ray images and show that the model tends to focus on the overall hand and only the bone part of the image, indicating its potential for accurately identifying the regions of interest for bone age assessment without the need for pre-processing to remove background noise. We also suggest two methods for extracting the region of interest from the heatmap output. Our study suggests that Vision Transformer holds great potential for bone age assessment using X-ray images, as it can provide accurate and interpretable output that may assist radiologists in identifying potential abnormalities or areas of interest in the X-ray image.

• Journal of the Institute of Electronics and Information Engineers
• /
• v.54 no.1
• /
• pp.80-87
• /
• 2017

This paper proposes a pre-processing algorithm to improve perceptual video coding efficiency which decomposes an input frame via a sub-band decomposition, and suppresses only high frequency band(s) having low visual sensitivity. First, we decompose the input frame into several frequency subbands by a linear sub-band decomposition. Next, high frequency subband(s) which is rarely recognized by human visual system (HVS) is suppressed by applying relatively small gain(s). Finally, the high frequency suppressed frame is compressed by a specific video encoder. We can find from the experimental results that if comparing before-use and after-use of the proposed pre-processing prior to the encoder, no visual difference is shown. Also, the proposed algorithm achieves bit-saving of 13.12% on average in a H.264 video encoder.

• Journal of Korea Multimedia Society
• /
• v.11 no.11
• /
• pp.1601-1614
• /
• 2008

The number of web service users has been increased rapidly as existing services are changed into the web-based internet applications. Therefore, it is necessary for us to use web log pre-processing technique to detect attacks on diverse web service transactions and it is also possible to extract web mining information. However, existing mechanisms did not provide efficient pre-processing procedures for a huge volume of web log data. In this paper, we proposed both a field based parsing and a high-speed log indexing mechanism based on the suggested B-tree Index Vector structure for performance enhancement. In experiments, the proposed mechanism provides an efficient web log pre-processing and search functions with a session classification. Therefore it is useful to enhance web attack detection function.

• Journal of Korean Society for Geospatial Information Science
• /
• v.23 no.1
• /
• pp.47-54
• /
• 2015

In this study, we developed strategies for pre-processing GNSS data for the purpose of separating geodetic factors from crustal deformation due to the earthquakes. Before interpreting GNSS data analysis results, we removed false signals from GNSS coordinate time series. Because permanent GNSS stations are located on a large tectonic plate, GNSS position estimates should be affected by the tectonic velocity of the plate. Also, stations with surrounding trees have seasonal signals in their three-dimensional coordinate estimates. Thus, we have estimated the location of an Euler pole and angular velocities to deduce the plate tectonic velocity and verified with geological models. Also, annual amplitudes and initial phases were estimated to get rid of those false annual signals showing up in the time series. By considering the two effects, truly geodetic analysis was possible and the result was used as preliminary data for analyzing post-seismic deformation of the Korean peninsula due to the Tohoku-oki earthquake.

• The Transactions of The Korean Institute of Electrical Engineers
• /
• v.66 no.2
• /
• pp.416-424
• /
• 2017

In this study, we propose a design of digits recognition system based on RBFNNs through a comparative study of pre-processing algorithms in order to recognize digits in handwritten. Histogram of Oriented Gradient(HOG) is used to get the features of digits in the proposed digits recognition system. In the pre-processing part, a dimensional reduction is executed by using Principal Component Analysis(PCA) and (2D)2PCA which are widely adopted methods in order to minimize a loss of the information during the reduction process of feature space. Also, The architecture of radial basis function neural networks consists of three functional modules such as condition, conclusion, and inference part. In the condition part, the input space is partitioned with the use of fuzzy clustering realized by means of the Fuzzy C-Means algorithm. Also, it is used instead of gaussian function to consider the characteristic of input data. In the conclusion part, the connection weights are used as the extended type of polynomial expression such as constant, linear, quadratic and modified quadratic. By using MNIST handwritten digit benchmarking database, experimental results show the effectiveness and efficiency of proposed digit recognition system when compared with other studies.

• The KIPS Transactions:PartB
• /
• v.10B no.1
• /
• pp.33-40
• /
• 2003

Data mining plays an important role in a knowledge discovery process and various algorithms of data mining can be selected for the specific purpose. Most of traditional hierachical clustering methode are suitable for processing small data sets, so they difficulties in handling large data sets because of limited resources and insufficient efficiency. In this study we propose a hybrid neural networks clustering technique, called PPC for Pre-Post Clustering that can be applied to large data sets and find unknown patterns. PPC combinds an artificial intelligence method, SOM and a statistical method, hierarchical clustering technique, and clusters data through two processes. In pre-clustering process, PPC digests large data sets using SOM. Then in post-clustering, PPC measures Similarity values according to cohesive distances which show inner features, and adjacent distances which show external distances between clusters. At last PPC clusters large data sets using the simularity values. Experiment with UCI repository data showed that PPC had better cohensive values than the other clustering techniques.

• Imaging Science in Dentistry
• /
• v.48 no.4
• /
• pp.261-268
• /
• 2018

Purpose: To determine the impact of an image processing technique on diagnostic accuracy of digital panoramic radiographs for the assessment of anatomical structures in paediatric patients with mixed dentition. Materials and Methods: The study consisted of 50 digital panoramic radiographs of children aged from 6 to 12 years, which were later on processed using a dedicated image processing method. A modified clinical image quality evaluation chart was used to evaluate the diagnostic accuracy of anatomical structures in maxillary and mandibular anterior and maxillary premolar region of processed images. Results: A statistically significant difference was observed between pre and post-processed evaluation of anatomical structures(P<0.05) in the maxillary and mandibular anterior region. The anterior region was found to be more accurate in post-processed images. No significant difference was observed in the maxillary premolar region (P>0.05). The Inter-observer and intra-observer reliability of both pre and post processed images were excellent (>0.82) for anterior region and good (>0.63) for premolar region. Conclusion: The application of image processing technique in digital panoramic radiography can be considered a reliable method for improving the quality of anatomical structures in paediatric patients with mixed dentition.

• Journal of the Korea Institute of Information and Communication Engineering
• /
• v.23 no.3
• /
• pp.276-281
• /
• 2019

Food printers must use food ink cartridges approved by the Ministry of Food and Drug Safety (MFDS). A 2D bar code is used to read whether the ink cartridge is authentic. However, since the dye is diverged by heat pressure and printed, the barcode is contaminated. In this paper, we propose a pre-processing algorithm to solve the problem of barcode contamination by food coloring dust in a latte art printer. The algorithm is based on various morphological operations. We apply this algorithm before reading contaminated barcode images with a general QR code reader. It has been confirmed that, as compared with the existing QR code reader, the contamination rate that can be perceived is increased from 25% to 40% and even at a contamination rate of 45%, the recognition rate reaches 50%.

• Journal of agriculture & life science
• /
• v.53 no.3
• /
• pp.147-157
• /
• 2019

In this study, we tried to find out the most appropriate pre-processing method and to verify the feasibility of developing a low-price sensing system for predicting the hardy kiwis sugar content based on VNIRS and subsequent spectral analysis. A total of 495 hardy kiwi samples were collected from three farms in Muju, Jeollabukdo, South Korea. The samples were scanned with a spectrophotometer in the range of 730-2300 nm with 1 nm spectral sampling interval. The measured data were arbitrarily separated into calibration and validation data for sugar content prediction. Partial least squares (PLS) regression was performed using various combinations of pre-processing methods. When the latent variable (LV) was 8 with the pre-processing combination of standard normal variate (SNV) and orthogonal signal correction (OSC), the highest R2 values of calibration and validation were 0.78 and 0.84, respectively. The possibility of predicting the sugar content of hardy kiwi was also examined at spectral sampling intervals of 6 and 10 nm in the narrower spectral range from 730 nm to 1200 nm for a low-price optical sensing system. The prediction performance had promising results with R2 values of 0.84 and 0.80 for 6 and 10 nm, respectively. Future studies will aim to develop a low-price optical sensing system with a combination of optical components such as photodiodes, light-emitting diodes (LEDs) and/or lamps, and to locate a more reliable prediction model by including meteorological data, soil data, and different varieties of hardy kiwi plants.

• Journal of Information Technology Services
• /
• v.19 no.5
• /
• pp.83-91
• /
• 2020

Language Models such as BERT has been an important factor of deep learning-based natural language processing. Pre-training the transformer-based language models would be computationally expensive since they are consist of deep and broad architecture and layers using an attention mechanism and also require huge amount of data to train. Hence, it became mandatory to do fine-tuning large pre-trained language models which are trained by Google or some companies can afford the resources and cost. There are various techniques for fine tuning the language models and this paper examines three techniques, which are data augmentation, tuning the hyper paramters and partly re-constructing the neural networks. For data augmentation, we use no-answer augmentation and back-translation method. Also, some useful combinations of hyper parameters are observed by conducting a number of experiments. Finally, we have GRU, LSTM networks to boost our model performance with adding those networks to BERT pre-trained model. We do fine-tuning the pre-trained korean-based language model through the methods mentioned above and push the F1 score from baseline up to 89.66. Moreover, some failure attempts give us important lessons and tell us the further direction in a good way.