Research And Implementation Of Video Based Motion De-blurring Technology

With the rapid development of the Mobile Internet technology and electronic information technology, all kinds of hand-held camera equipment has been widely spread and people are more and more inclined to use video to record their daily life. However, in the process of real time shooting through hand-held cameras, the relative motion between cameras and objects, which is caused by camera shakes or movement of objects existing in the target scene, will lead captured video to occur motion blur, which will cause a serious loss of high frequency details and thus reduce the video quality. According to the above application background, in this paper, a new gray distribution characteristics and interpolation based motion video de-blurring algorithm is proposed.For the shortcomings of existing methods, the proposed gray distribution based feature description method selects the feature point as the center of a circular neighborhood window, by calculating the gray distribution within the eight equal fan-shaped area, which starts from the major orientation to the feature point, to acquire the corresponding component values of feature vector for the matching operation of subsequent processing algorithms. Finally, the comparison and test of objective data and subjective result is given to evaluate the performance of our feature description algorithm. The experimental results show that our algorithm is invariant to rotation and compared to Hu moment and its improved feature description method, our algorithm can maintain relative higher matching accuracy and has a more stable processing performance and stronger robustness.In order to restore motion blurred videos shoot in the real world, this paper presents an improved interpolation based motion video de-blurring algorithm. Our method adopts the inverse of the sum of squared gradient measure computed in a common coverage area which is observed in the neighboring frames to evaluate the relative blurriness. Meanwhile, by taking full account of the information of all pixels in the space-time neighborhood window of processing pixel, classic Laplace operator is used to get the corresponding component values for weighted operation in the neighborhood window of registration pixels. Finally, the above component values for weighted operation is multiplied by the weighted value, which is the ratio of the measure of the two frames where registration pixels is located, to calculate the interpolation values to replace the current processing pixel. Experimental results show that our algorithm can achieve a magnificent motion de-blurring performance and have a full capability to retain the edge information of the processing video.Finally, combined with the demand of practical applications, our motion video de-blurring system, which takes advantage of the simple and high efficiency of our imposed algorithm, is implemented on OpenCV platform and achieve an excellent performance.