A new algorithm, CBIR, has been developed to improve the resolution of image registration for PSP. The displacement vectors for the local motion of the model were obtained by calculating the maximum cross-correlation between the 'wind-off and 'wind-on' images. A layer of speckle was sprayed to apply the cross-correlation method in a small interrogation window, without requiring the identification of control points as used in the conventional method. Recursive multigrid processing was employed to increase the nonlinear spatial resolution. It was found that high-resolution images could be obtained by CBIR without requiring control points. The relative-error distributions and the mean relative error (MRE) were employed to evaluate the image-registration methods. The conventional method with control points showed a relatively small error, although large errors were produced in the boundary region and in the vicinity of the control points, where steep intensity changes occur. The present CBIR showed relatively small errors and described a nonlinear deformation accurately without control points. Of the interpolation methods evaluated for the image transformation procedure, windowed-sinc interpolation was the most accurate and 4x4 cubic interpolation was the most efficient.
Accuracy of correlation-based image registration (CBIR) for pressure-sensitive paint was investigated. In the present study, the influence of displacement errors and their sensitivity on accuracy of pressure measurement was described with an uncertainty analysis. The error sources of image registration were classified and several factors affecting the accuracy of image registration were examined. The performances of image registration were compared under several artificial model motions. The local intensity variation due to speckles, which enhance the image correlation in CBIR, could act as a source of image noise and deteriorate the local pressure sensitivity. The local pressure sensitivity with speckles was investigated through pixel-by-pixel calibration. A spatial filtering was employed to reduce the local intensity variation. It was found that the fluctuations of the local pressure sensitivity are decreased and signal-to-noise ratio is improved with a median filter.
압력감응 페인트(pressure-sensitive paint, PSP)를 사용하여 압력을 측정하는 방법 중 가장 많이 사용되는 방법은 '유동이 없는' 상태의 PSP의 밝기와 '유동이 있는' 상태의 PSP의 밝기의 비를 통해 압력장을 구하는 방법이다. 하지만 유동이 있는 상태의 이미지와 유동이 없는 상태의 이미지 사이에 움직임이 있으면 오차가 발생하기 때문에 정확한 압력 값을 얻어내기 위해서는 유동이 있는 이미지의 움직임을 보정해 주는 작업인 이미지 등록(image registration)이 필수적이다.
본 연구에서는 기존의 이미지 등록의 단점을 개선하기 위한 새로운 이미지 등록방법을 제안하였다. 상관법에 의한 이미지 등록법은 wind-off와 wind-on 두 이미지 사이의 상관함수을 계산하여 각 조사구간의 변위를 구함으로써 유동에 의한 모형의 움직임을 보정하는 방법이다. 두 이미지의 작은 조사구간에서 상관관계를 극대화하기 위해서 회색의 얼룩무늬를 분무하였다. 상관법에 의한 이미지 등록법을 충돌제트 유동장에 적용하여 본 결과, 기준점의 도움 없이 기존의 이미지 등록보다 정밀한 압력측정이 가능하다는 것을 확인하였다.
상관법에 의한 이미지등록의 정확도를 정량적으로 해석해 본 결과, 회전과 굽힘 변형에 대해 CBIR은 전반적으로 0.1픽셀이하의 오차를 보이며 다항함수에 의한 이미지등록보다 좋은 성능을 나타내었다. 그리고 중간값 필터를 통한 스페클 제거효과와 변위오차에 대한 민감도 감소를 확인하였다.