| Due to the rapid development of multimedia technology, various videos with different frame rates have appeared, and the frame rate conversion of such videos is inevitable. Frame rate up conversion (FRUC) is used for the above situation, and it aims to convert from a lower frame rate to a higher one. Besides, with people’s demands of high-quality videos, the applications of high-definition (HD) videos are more and more common. Thus, the research on FRUC for HD videos is necessary. At present, the most popular applications of FRUC include:reducing the motion judder and motion blur in liquid crystal display (LCD) when displaying moving scenes, restoring the original frame rate of the low frame rate video system, slow motion playback, scalable video coding, and so on.Most of the current FRUC methods are based on motion compensation, which can achieve smooth visual effect. The motion vectors used in motion compensation can be transmitted from the encoder side, or can be re-estimated at the decoder side. This thesis researches the FRUC algorithm which re-estimates the motion vectors, and the FRUC algorithm can be applied in HD video system.The contents of the research in this thesis are as following:1. Motion estimation structures include unidirectional motion estimation and bilateral motion estimation. Based on the unidirectional motion estimation, a forward and backward jointing motion estimation method is proposed. The proposed motion estimation method combines forward motion estimation and backward motion estimation, and improves the accuracy of the motion estimator. Based on the motion estimator, an FRUC algorithm called FBFRC (Frame Rate Conversion Based on Forward and Backward Motion Estimation) is presented.2. In the FBFRC scheme, a motion vector field retiming method based on median filter is presented. The motion vector field retiming is used to generate the motion vector fields of the to-be-interpolated frames. The proposed median filter based motion vector field retiming method avoids the holes and overlapped problems in the traditional interpolation process.3. In the FBFRC scheme, an occlusion detection method is presented. The proposed method divides the image into covered regions, exposed regions, and non-occludded regions through analyzing all the motion types of two adjacent objects, and achieves the motion models of the above three regions. Based on the results of the above occlusion detection, unidirectional interpolation is adopted for the covered and the exposed regions, and bidirectional interpolation is used for the non-occlusion regions, which effectively solves the halo artifacts around the objects without occlusion handling.4. Two motion estimation methods based on bilateral motion estimation are proposed, which are frequency and spatio-temporal motion estimation and a bilateral motion estimation method. The frequency domain motion estimation adopts phase plane correlation. The accuracy of the motion has been improved through combining the phase plane correlation with spatio-temporal motion estimation. The proposed bilateral motion estimation improves the accuracy and the consistency of the motion estimation by using an adaptive update strategy and a motion vector constraint. Based on the above two motion estimation methods, two FRUC algorithms are presented separately, which are FSTFRC (Frame Rate Conversion Based on Frequency and Spatio-Temporal Motion Estimation) and BLFRC (Frame Rate Conversion Based on Bilateral Motion Estimation).5. In the BLFRC scheme, a motion vector field smoothing method is proposed. It uses the localized global motion in the neighborhood of the current block, and the localized global motion is produced by motion vector histogram. The motion vector smoothing corrects the false motion vectors generated in the motion estimation process, which enhances the consistency of the motion fields and improves the visual quality of the interpolated frames.6. Three proplems met in the applications of FRUC have been analyzed, and this thesis gives the solutions.(1) If the input video source is film mode, the input video contains lots of repetitive frames. There will be more repetitive frames in the output video of the FRUC process, and the audience will feel jitter when watching the video. The above problem can be solved by removing the repetitive frames before FRUC through film mode detection.(2) When the input video contains texts, the motion vectors of the text regions may be affected by the motion vectors of the surrounding non-text regions, which results in false motion estimation and quality degradation of the interpolated frames. To solve the above problem, a scrolling text processing method and a static text detection method are proposed. The proposed schemes increase the motion estimation accuracy of the text regions, and improves the quality of the interpolated frames of the video with texts.(3) When scene cut occurs in the input video, motion estimation between the adjacent two frames will generate false motion vectors, and the motion compensated frame rate up conversion will produce seriously distorted interpolation results. To solve the above problem, a scene cut detecion scheme based on the matching error is presented in this thesis. When scene cut is detected, motion compensation is replaced by frame repetition or smooth frame insertion to interpolate the frames, and better visual quality can be achieved. |
PDF Full Text Request