In this thesis, a 3D projection profile-based content aware video retargeting scheme is proposed. The 3D projection profiles use the horizontal, vertical, and time directions to project the energy features for video spatial and temporal resizing. Recently, a new technique known as seam carving is proposed to resize images or videos to fit different aspect ratios. This method computes an energy map by extracting image gradient features and finds out 1D seams or 2D seam manifolds with minimal energy. By removing or inserting the 1D seams or 2D seam manifolds, the image or video can be adjusted to fit the target aspect ratio and size. However, the main problem of this method is the structure misalignment or shape destruction in the resultant images or videos, e.g. discontinuous, twisted, distorted, or broken objects. The method in this thesis is proposed to solve this problem. First, an energy map which represents the importance content of each frame is calculated and combined by a saliency-based visual attention map and a motion detection map. Then the scheme uses the 2D projection profiles of the 3D video energy volumes along horizontal, vertical, and time axes to obtain important video content. The projection profiles in 2D are used to find out the carving seams for video spatial retargeting and cutting frames for video temporal resizing. By this way, the discontinuity and misalignment that usually occur in the results of video seam carving are reduced. Experimental results show that the proposed method is capable of yielding more acceptable video quality.
