Picture Quality Editing for Variable Rate Compressed Video

Abstract

This paper discusses a novel video encoding system in which a user controls the quality of the encoded video. It uses an algorithm for reallocation of target bits to the segment selected for editing or quality change. The algorithm is based on the availability of a log file containing statistics pertaining to the original run that produced the compressed video. A key aspect of the bit allocation is that subjective input from the user is used to alter the picture quality of the edit segment. Bitstream splicing is carried out to complete the process of editing. The video is initially encoded according to an automatic process without user intervention. An operator then reviews it manually and indicates the quality of specific time periods or regions to be altered. A numeric value for picture quality is assigned in the range [-5, 5]. First, a common bit pool is created by removing a percentage of bits from the time periods or regions for which quality priorities have been specified. These are then redistributed as a function of the user-selected priority. The bit assignments are adjusted resulting in the avoidance of buffer underflow. Also, the buffer occupancy at the trailing edge of the edit segment is maintained at its original level. After the assignment of new target bits, the average quantizer scale value is determined using a previously determined rate quantizer function. The video is then re-encoded with the new set of target bits to achieve the desired quality change. Bitstream and log files are updated by proper splicing.