| Popis: |
The 3D-DCT is used in video compression when the applied basic principle is the correlation between adjacent pixels in a frame and the correlation between pixels from the same position in adjacent frames (temporal correlation). This principle uses the concept that a video sequence can be seen as a parallelepiped representing the entire video sequence, divided into smaller, independently encoded parallelepipeds, the most commonly format being a "video cube" of 8×8×8 size. When passing this cube through the 3D-DCT algorithm that applies to it, the corresponding quantization tables must be known by both the encoder and the decoder to be able to decompress the video information. This paper presents an original method of implementing a modified 3D-DCT compression algorithm, which is a variant derived from it and which brings certain improvements to the standard algorithm. The main improvement is to replace fixed-sized "video cubes" by creating variable-length parallelepipeds over time, giving up the fixed value of 8 for the "video cube" dimension, which means considerable reductions in the volume of this parallelepiped when its sides are smaller than 8. n-bit operations can be performed more efficiently and resource-saving when n < 8 and implicitly the value of 2n – 1 no longer complies with the divisibility by 8. Other improvements are related to how the information bits are saved when storing them in the compressed video file. Reading from this file and writing in this file run in large blocks, to avoid accessing the file too often with short readings or writings. The design and implementation details, as well as the analysis of the compression performance obtained (including the consumption of time divided by type of function used) are thoroughly presented in the paper. |
