DCT-V for Video Coding: A reconfigurable implementation for length 32 and 4.

This work of thesis shows a new factorization and its implementation for DCT-V(Discrete Cosine Transform) of length 4 and 32, used in the most recent video compression standard. During the last years, the Joint Video Exploration Team(JVET) of ITU-T VCEG and ISO/IEC MPEG, now the Joint Video Experts Team, are trying to develop the future video coding technology, that will improve the compression efficiency with respect to the current HEVC standard. An experimental software was introduced, JEM, based on the HEVC Model (HM) software, developed as reference for the HEVC standard. One of the fundamental data compression techniques used for this standard, as in many previous image and video coding standards, is transform coding. What is interesting to consider for the future video coding is the introduction of the new transforms like DCT-V, DCT-VIII, DST-I and DST-VII. This new transforms need low-complexity factorization to efficiently compute them. This work exploits a factorization of the DCT using the algebraic theory, the Reznik decomposition for DCT-II and an already known fast algorithm for the DCT-V of length 4. In particular, all this known relationships are combined together to find that DCT-V of length 32 can be computed using 5 of the same type DCTs of length 4. Moreover, a Matlab model has been written for functional validation, followed by a model written in C language. This allows for the evaluation of the rate-distortion performance within the video codex for the new DCT-V of length 32 and 4. To do this, it is important to adapt the proposed algorithms for DCT-V with the existing DCT-V description in the JEM software, that is a normalized version with integer coefficients. Finally a reconfigurable unit exploiting multiplexers has been designed, which computes both the DCT-V of length 32 and 5 DCTs of the same type of length 4. For synthesis the umc65 library is used and the reports of time, area and power are obtained.