Development of Fast Prallel Algolithms for Stochastic Moving Image Processing and its Application in Image Measurements
Grant-in-Aid for General Scientific Research (B)
|Allocation Type||Single-year Grants|
|Research Institution||KYOTO UNIVERSITY|
OGURA Hisano Kyoto Univ., Graduate School of Engineering, Professor, 工学研究科, 教授 (50025954)
TAKAHASHI Nobuyuki Univ.Shiga Prefecture, International Education Center, Associate Professor, 国際教育センター, 助教授 (70206829)
MIYAGI Shigeyuki Kyoto Univ., Graduate School of Engineering, Instructor, 工学研究科, 助手 (20273469)
|Project Period (FY)
1994 – 1995
Completed(Fiscal Year 1995)
|Budget Amount *help
¥6,400,000 (Direct Cost : ¥6,400,000)
Fiscal Year 1995 : ¥2,200,000 (Direct Cost : ¥2,200,000)
Fiscal Year 1994 : ¥4,200,000 (Direct Cost : ¥4,200,000)
|Keywords||whiteness / noncausal model / drift moving image / orthogonalization method / spectral estimation / texture analysis / target detection / wavy random image / 定常動画像 / 白色化 / 空間フィルタ / 実動画像処理 / 非因果性画像モデル / パラメタ推定 / 時空間白色化フィルタ / 並列アルゴリズム / 確率過程|
We have developed the method of estimating a stochastic model for a random moving image which can be regarded as a homogeneous and stationary random field in 3D space, and the method was applied to variety of image processings.
(1) Application to the modeling of a drift moving image
A method of estimating a stationary moving image was developed by combining the method of estimation for a spatial noncausal model with the one for an temporal AR model used in the time series analysis. We successfully applied this technique to the model estimation of a drifting, moving image also.
(2) Proposal of the quasi-sequential orthogonalization method In the processing of real moving images it is sometimes insufficient for whitening a real moving image to use only the method of noncausal spatial modeling combined with temporal AR modeling. To improve this, the method of sequential orthogonalization, developed in the time series analysis, has been applied to the moving image processing, and it was made
possible to get an almost perfect whitening filter by this method for the space-temporal modeling. When the whitening is insufficient, we can repeat the quasi-sequential orthogonalization to get a better whitening filter for the modeling.
(3) The spectral estimation of a real moving image
Above method was applied to the spectral estimation of a real, moving image. In the processing of a drifting, moving image, it was also shown that the drift velocity can be estimated making use of the spatial correlation functions.
(4) Applications to the texture analysis
It is shown that the noncausal model filter for a still or moving image can be applied to detecting a target and to detecting the border of neighboring textures. As a result, S/N ratio is improved by the processing of moving image.
(5) The model estimation for an image with double peaked spectrum
A new attempt was made to estimate the model for an image with double peaked spectrum. The model estimation is made possible for a wavy random image having double spectral peaks, and the new method can also be applied to wavy random moving image. Less
Research Output (11results)