摘要
在曝光的瞬间,造成图像模糊的被摄景物与相机的相对运动,虽可近似作为直线运动来处理,但模糊图像中的运动模糊方向未知。若能由模糊图像出发,估计出运动模糊方向,则可以通过图像旋转将运动模糊方向旋转到水平轴,这样图像恢复就可由2维问题转化为1维问题,这就大大降低了图像恢复的难度,并为图像恢复的并行计算创造了有利条件。为实现这一目的,将原图像看作是各向同性的一阶马尔科夫过程,提出了一种新的基于方向微分的运动模糊方向鉴别方法,该方法不仅可以高精度鉴别匀速运动、加速运动、振动等各种运动的模糊方向,而且具有鉴别范围大、鉴别精度高、稳定性好的优点。另外,为了具体实现这种鉴别,还给出了采用双线性插值或三次C样条插值进行方向鉴别的详细计算方法,其中双线性插值方法计算量小,但三次C样条插值方法的鉴别精度比双线性插值方法高,而且通过加权平均,还可进一步降低各种随机因素引起的鉴别误差,这不仅提高了鉴别精度,而且增强了运动模糊方向鉴别的稳定性,因此能够更加有效地进行方向鉴别。
The direction of the motion that blurs the image can be dealt with as unchanged during the short expose time. It is a pity that the actual motion blur direction in the blurred image is often unknown. It is meritorious to identify the motion blur direction from the blurred images. Then the motion blurred direction can be adjusted to the horizontal axis by image rotation, and the image restoration will become a one-dimension problem, which is much more easy than the two-dimension that it was. Dealing with the original image as an isotropy one rank Markov process, this paper give a new way to identify the motion blur direction from the blurred images by direction derivation method. The new way can identify the blur directions of the images blurred by uniform motion, accelerate motion, vibration etc. It can identify any direction, from -90° to 90°,with high precision. It works with high stabilization.Double linear interpolation or C spline interpolation is used in the identification. The detailed computing processes of the interpolations are presented. Double linear interpolation needs less computing time, but C spline interpolation works out with higher identification precision. The new way to identify the motion blur direction is developed by introducing into the weighted averages method, which is helpful to restrain the random factors that cause identification errors, to enhance the identification stabilization, to work out with higher identification precision.
出处
《中国图象图形学报（A辑）》
CSCD
北大核心
2005年第5期590-595,共6页
Journal of Image and Graphics
