|
Number Recognition of Parts Book Schematics using Convolutional Recurrent Neural Network |
|
Erdal Genc, Hee Ran Shin, Jang Sik Park, Jong Kwan Song(Kyungsung Univ., Korea) |
| First, is the enhancement of the device to operate more productively even if the number of employees is decreased. Secondly, is the increase in the efficiency of the storage. This paper compares two state-of-the-art OCR algorithms in a simulated environment by using modified dataset. |
|
|
Contour Segmentation Based on Density Gradient and Region Growing |
|
Yang Yu, Kang-Hyun Jo(University of Ulsan, Korea) |
| This paper proposes a 3D vehicle contour segmentation method based on density gradient and region growing. The ground data are removed by RANSAC, then the 3D point cloud projected to the 2D grid. The density gradient is used to detect the edge points in grid. Then the vehicle contour is segmented by region growing method. The vehicle contour segmentation result has smoother representation and less noise, so this method has a good practical value. |
|
|
Clustering for Electronic Warfare Information |
|
Eren Yıldırım, Yucel Batu Salman(Bahcesehir University, Turkey), Jang Sik Park(Kyungsung University, Korea) |
| Information is collected through different methods such as electronic warfare. Electronic warfare is a set of activities using electromagnetic spectrum in favor of the host. This study aims to analyze the usage of datasets gathered through electronic warfare to determine regions of target elements through DBSCAN and K-Means. Time series analysis will reveal the coordinates of the opponent forces according to communication frequency. Lastly, by using statistical data, a study is made to reach important information like widely used frequencies, frequency, route maps and displacements. |
|
|
Reduction of background sound based on non-negative matrix factorization combined with Wiener filter post-processing |
|
Jang Sik Park(Kyungsung University, Korea) |
| In this paper, we propose a method to reduce background noise mixed with speech by using Nonnegative Matrix Factorization. The proposed method is a method of suppressing the background sound component once more through the Wiener filter in the time domain based on the speech component separated or reduced from the background sound mixed voice by the Nonnegative Matrix Factorization method. Experimental results show that the proposed method is effective. |
|