Motion Compensation Frame Rate Up-conversion Algorithm Based On Visual Feature Extraction

With the increasing popularity of network,video has been widely used to spread information,and video communication equipment has imperceptibly entered people's life.However,it is difficult to accommodate a lot of video data and get video resources because of the limited storage and processing capacity of video communication equipment,so the video cannot be played smoothly.Meanwhile,due to the rapid development of network information technology,the clarity and fluency of current video can no longer meet people's visual needs.Therefore,how to improve the quality of video has become the focus of research.Video communication equipment reduces the video frame rate by dropping frames when transmitting video to provide high quality video.Frame rate up-conversion technology is then used to reconstruct the lost video frames at the video receiver,making frame rate up-conversion technology widely used in the multimedia industry.This dissertation focuses on enhancing the accuracy of motion estimation in frame rate up-conversion based on relevant theories,and extracting artificial and machine vision features to boost video quality in order to produce high-quality video.The following is a list of the specific research topics covered in this dissertation:(1)On the basis of the context cube features,a motion compensation frame rate up-conversion technique is proposed.The proposed algorithm uses the extracted context cube features to perform bidirectional context cube matching and estimate the motion vectors of the interpolated frames.Then motion compensation interpolation is implemented to obtain the interpolated frames.When compared to conventional algorithms,the proposed algorithm can provide better subjective and objective visual effects.(2)On the basis of the context statistical features,a motion compensation frame rate up-conversion technique is proposed.Based on the context cube features,the proposed algorithm further extracts the context statistical features,that is,the features of the reference frame is represented by the expectation or skewness.The extracted context statistical features is fused with the original frame,so as to achieve bidirectional context statistical feature matching,and generate the motion vector field of the intermediate frame.The proposed algorithm can effectively reduce the time complexity.For most video sequences,the quality of the interpolated frames produced by this algorithm is superior to that of the motion compensation frame rate up-conversion algorithm based on context cube features.(3)To further compare the performance of artificial and machine features,a motion compensated frame rate up-conversion algorithm based on machine vision features is proposed.The proposed algorithm uses the improved pre-training neural network AlexNet to extract the features of each reference frame,and combines the extracted features with the original reference frame,then the bidirectional AlexNet feature matching is implemented to obtain the motion vectors of the interpolated frame.According to experimental results,compared with the traditional motion compensation frame rate up-conversion algorithm,the proposed algorithm can improve the quality of the interpolated frames to a certain extent,and the proposed artificial vision features are compared with AlexNet machine vision features.It can be seen that the machine vision features based on AlexNet have similar performance to the proposed artificial vision features.Therefore,the proposed artificial vision features have certain competitiveness.