Papers - THI THI ZIN
-
Accurate estimation of missing data under noise distribution Reviewed
S-S. Koh, Thi Thi Zin, H. Hama
IEEE Transactions on Consumer Electronics 52 ( 2 ) 528 - 535 2006.5
Language：English Publishing type：Research paper (scientific journal) Publisher：IEEE Transactions on Consumer Electronics
3D contents have been becoming one of attractive multimedia and reality. In computer vision, 2D-to-3D conversion techniques require estimating missing data from noisy observations. When there is no missing data in the observation matrix, the accurate solution of such problem is known to be given by Singular Value Decomposition (SVD). In the case of converting already recorded monoscopic video contents to 3D, several entries of the matrix have not been observed. Therefore, the problem has no simple solution, so it is necessary to estimate missing data. In this paper, we propose an estimation algorithm of missing data with minimizing the influence of noise embedded when tracking feature points from partial observations. The proposed method is an iterative affine SVD factorization method which can estimate the model parameters, given an incomplete set of the observation matrix. The main idea of our algorithm is to estimate missing data accurately even under noise distribution by using geometrical correlations between 2D and 3D error space. This paper consists of three main phases: geometrical correlations for estimating missing data, estimation algorithm, and analyzing the results for video sequences. The accurate results in practical situations as demonstrated here with synthetic and real video sequences show the efficiency and flexibility of the proposed method.
-
Relative color polygons for object detection and recognition
Zin T., Koh S., Hama H.
Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics) 4259 LNAI 834 - 843 2006
Language：Japanese Publishing type：Research paper (scientific journal) Publisher：Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
This paper proposes a new framework of the color model for outdoor scene image detection and recognition. This model enables us to manipulate easily the color of an image. Here, the concept of 'relative color polygon' for an object composed of uniform color regions is introduced on a 2D color space (XY space). Then the color similarity is defined using three kinds of parameters of the polygon: length and slope of every side and angle of adjacent sides. This paper addresses how to decide the color similarity by using the facts about color shifting on the XY space. The feasibility of the proposed framework has been confirmed through the experimental results using outdoor scene images taken under a great variety of various illumination conditions. © Springer-Verlag Berlin Heidelberg 2006.
-
Robust segmentation of road traffic signs using adaptive thresholds
Liao C.
IEICE Electronics Express 2 ( 14 ) 423 - 428 2005.11
Language：English Publishing type：Research paper (scientific journal) Publisher：IEICE Electronics Express
We propose a robust segmentation system for road traffic signs using adaptive thresholds. The main objective is to develop a robust segmentation system under various illumination conditions. The main problem is how to obtain the best suited threshold for relative similarity. It can be got automatically by K-nearest connectivity when K = 4 and the two-way connectivity method. Our system utilizes relative similarity of neighborhood pixels and shape information of traffic signs together. The effectiveness of our proposed system is confirmed through experiments under various illumination conditions. In the ex-periments, 90 images taken under daytime, evening and night-time are used. The system can give 98% successful segmentation rate. © 2005, The Institute of Electronics, Information and Communication Engineers. All rights reserved.
DOI： 10.1587/elex.2.423
-
A robust road sign recognition using segmentation with morphology and relative color Reviewed
Thi Thi Zin, H. Hama
映像情報メディア学会 59 ( 9 ) 1333 - 1342 2005.9
Language：English Publishing type：Research paper (scientific journal) Publisher：Kyokai Joho Imeji Zasshi/Journal of the Institute of Image Information and Television Engineers
We propose a robust road sign recognition system under various illumination conditions. The proposed approach has two steps: segmentation and recognition. The segmentation, which is the focus of this paper, is performed using morphological operations and relative color. The segmented regions are recognized by a template matching method using modified standard deviation. The algorithm works for various types of circular and pentagonal road signs. In experiments under various illumination conditions, the segmentation rate was 100% in the daytime and evening and 80% even in the night-time, and the recognition rate was 100% for all of the segmented regions under all illumination conditions. The effectiveness of the proposed system was confirmed through experiments using 200 images of road signs taken under a great variety of illumination conditions including fog and light rain.
-
Robust road sign recognition using standard deviation
Zin T., Hama H.
IEEE Conference on Intelligent Transportation Systems, Proceedings, ITSC 429 - 434 2004.12
Language：Japanese Publishing type：Research paper (scientific journal) Publisher：IEEE Conference on Intelligent Transportation Systems, Proceedings, ITSC
This paper proposes the recognition of road signs under various illumination conditions using standard deviation. The proposed road sign recognition system uses both of shape and color information. The former one is used in template matching with standard deviation and the latter one in modified Mahalanobis distance. This paper focuses on circular road signs and achieves almost perfect performances under various conditions such as daytime, evening-time and nighttime. Moreover, the proposed method may be applicable to recognize of any shape of road signs. The effectiveness of the proposed method has been demonstrated through experiments. According to our experimental results, the recognition rate is 100% for 200 images taken under various illumination conditions.