The frame skipping at transmitter and its reconstruction by interpolation at receiver is an essential process for low bit rate video transmission. However, simple interpolation method has several problems such as blurring and jerkiness. To solve this problem, motion compensated interpolation (MCI) schemes using segmentation and motion information are used. The accuracy of the motion information is very important for the performance of MCI.
In this thesis, a new MVC (Motion Vector Correction) technique is proposed for use in block based motion estimation. This approach consists of two steps. The first step decides the agreement of the actual motion of object and the estimated motion vector using the relation with adjacent motion vectors. The second step eliminates the non-uniformity of motion vectors which is originated from motion vectors with more or less different direction. Since the proposed MVC scheme corrects the estimated motion vector using local characteristics of an image, the MCI using this MVC is effective in real image.
With the MCI using the proposed MVC, we can reduce the degradation of picture quality in the interpolated image. Especially, if this method is applied to images with various motions, we can obtain better picture quality than the existing method.